[ { "id": "https://authors.library.caltech.edu/records/60m6h-a7p32", "eprint_id": 120107, "eprint_status": "archive", "datestamp": "2023-08-20 16:40:31", "lastmod": "2023-10-25 16:52:46", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Voloshin-Cameron", "name": { "family": "Voloshin", "given": "Cameron" } }, { "id": "Verma-Abhinav", "name": { "family": "Verma", "given": "Abhinav" }, "orcid": "0000-0002-9820-8285" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Eventual Discounting Temporal Logic Counterfactual Experience Replay", "ispublished": "unpub", "full_text_status": "public", "note": "Attribution 4.0 International (CC BY 4.0)\n\n
Submitted - 2303.02135.pdf
", "abstract": "Linear temporal logic (LTL) offers a simplified way of specifying tasks for policy optimization that may otherwise be difficult to describe with scalar reward functions. However, the standard RL framework can be too myopic to find maximally LTL satisfying policies. This paper makes two contributions. First, we develop a new value-function based proxy, using a technique we call eventual discounting, under which one can find policies that satisfy the LTL specification with highest achievable probability. Second, we develop a new experience replay method for generating off-policy data from on-policy rollouts via counterfactual reasoning on different ways of satisfying the LTL specification. Our experiments, conducted in both discrete and continuous state-action spaces, confirm the effectiveness of our counterfactual experience replay approach.", "date": "2023-03-17", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20230316-204049328", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20230316-204049328", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "primary_object": { "basename": "2303.02135.pdf", "url": "https://authors.library.caltech.edu/records/60m6h-a7p32/files/2303.02135.pdf" }, "resource_type": "monograph", "pub_year": "2023", "author_list": "Voloshin, Cameron; Verma, Abhinav; et el." }, { "id": "https://authors.library.caltech.edu/records/0gsec-n8613", "eprint_id": 118474, "eprint_status": "archive", "datestamp": "2023-08-20 08:41:54", "lastmod": "2023-10-24 23:22:35", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Huang-Yujia", "name": { "family": "Huang", "given": "Yujia" }, "orcid": "0000-0001-7667-8342" }, { "id": "Jimenez-Rodriguez-Ivan-Dario", "name": { "family": "Jimenez Rodriguez", "given": "Ivan Dario" }, "orcid": "0000-0001-9065-5227" }, { "id": "Zhang-Huan", "name": { "family": "Zhang", "given": "Huan" }, "orcid": "0000-0002-1096-4255" }, { "id": "Shi-Yuanyuan", "name": { "family": "Shi", "given": "Yuanyuan" }, "orcid": "0000-0002-6182-7664" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "FI-ODE: Certified and Robust Forward Invariance in Neural ODEs", "ispublished": "unpub", "full_text_status": "public", "note": "This work is funded in part by AeroVironment and NSF #1918865.", "abstract": "We study how to certifiably enforce forward invariance properties in neural ODEs. Forward invariance implies that the hidden states of the ODE will stay in a \"good\" region, and a robust version would hold even under adversarial perturbations to the input. Such properties can be used to certify desirable behaviors such as adversarial robustness (the hidden states stay in the region that generates accurate classification even under input perturbations) and safety in continuous control (the system never leaves some safe set). We develop a general approach using tools from non-linear control theory and sampling-based verification. Our approach empirically produces the strongest adversarial robustness guarantees compared to prior work on certifiably robust ODE-based models (including implicit-depth models).", "date": "2022-12-21", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20221219-234122405", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20221219-234122405", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "AeroVironment" }, { "agency": "NSF", "grant_number": "CCF-1918865" } ] }, "doi": "10.48550/arXiv.2210.16940", "resource_type": "monograph", "pub_year": "2022", "author_list": "Huang, Yujia; Jimenez Rodriguez, Ivan Dario; et el." }, { "id": "https://authors.library.caltech.edu/records/63nae-74q84", "eprint_id": 118473, "eprint_status": "archive", "datestamp": "2023-08-20 08:38:08", "lastmod": "2023-10-24 23:22:32", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Sun-Jennifer-J", "name": { "family": "Sun", "given": "Jennifer J." }, "orcid": "0000-0002-0906-6589" }, { "id": "Tjandrasuwita-Megan", "name": { "family": "Tjandrasuwita", "given": "Megan" } }, { "id": "Sehgal-Atharva", "name": { "family": "Sehgal", "given": "Atharva" } }, { "id": "Solar-Lezama-Armando", "name": { "family": "Solar-Lezama", "given": "Armando" }, "orcid": "0000-0001-7604-8252" }, { "id": "Chaudhuri-Swarat", "name": { "family": "Chaudhuri", "given": "Swarat" }, "orcid": "0000-0002-6859-1391" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Costilla-Reyes-Omar", "name": { "family": "Costilla-Reyes", "given": "Omar" }, "orcid": "0000-0001-8331-7262" } ] }, "title": "Neurosymbolic Programming for Science", "ispublished": "unpub", "full_text_status": "public", "note": "Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).\n\nThis project was supported by the National Science Foundation under Grant #1918839 \"Understanding the World Through Code\" http://www.neurosymbolic.org/\n\nAccepted Version - 2210.05050.pdf
", "abstract": "Neurosymbolic Programming (NP) techniques have the potential to accelerate scientific discovery. These models combine neural and symbolic components to learn complex patterns and representations from data, using high-level concepts or known constraints. NP techniques can interface with symbolic domain knowledge from scientists, such as prior knowledge and experimental context, to produce interpretable outputs. We identify opportunities and challenges between current NP models and scientific workflows, with real-world examples from behavior analysis in science: to enable the use of NP broadly for workflows across the natural and social sciences.", "date": "2022-12-21", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20221219-234119032", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20221219-234119032", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NSF", "grant_number": "CCF-1918839" } ] }, "doi": "10.48550/arXiv.2210.05050", "primary_object": { "basename": "2210.05050.pdf", "url": "https://authors.library.caltech.edu/records/63nae-74q84/files/2210.05050.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Sun, Jennifer J.; Tjandrasuwita, Megan; et el." }, { "id": "https://authors.library.caltech.edu/records/z042w-bx383", "eprint_id": 118472, "eprint_status": "archive", "datestamp": "2023-08-20 08:21:38", "lastmod": "2023-10-24 23:22:30", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Tucker-Maegan", "name": { "family": "Tucker", "given": "Maegan" }, "orcid": "0000-0001-7363-6809" }, { "id": "Li-Kejun", "name": { "family": "Li", "given": "Kejun" }, "orcid": "0000-0002-0823-9839" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Ames-A-D", "name": { "family": "Ames", "given": "Aaron D." }, "orcid": "0000-0003-0848-3177" } ] }, "title": "POLAR: Preference Optimization and Learning Algorithms for Robotics", "ispublished": "unpub", "full_text_status": "public", "abstract": "Parameter tuning for robotic systems is a time-consuming and challenging task that often relies on domain expertise of the human operator. Moreover, existing learning methods are not well suited for parameter tuning for many reasons including: the absence of a clear numerical metric for `good robotic behavior'; limited data due to the reliance on real-world experimental data; and the large search space of parameter combinations. In this work, we present an open-source MATLAB Preference Optimization and Learning Algorithms for Robotics toolbox (POLAR) for systematically exploring high-dimensional parameter spaces using human-in-the-loop preference-based learning. This aim of this toolbox is to systematically and efficiently accomplish one of two objectives: 1) to optimize robotic behaviors for human operator preference; 2) to learn the operator's underlying preference landscape to better understand the relationship between adjustable parameters and operator preference. The POLAR toolbox achieves these objectives using only subjective feedback mechanisms (pairwise preferences, coactive feedback, and ordinal labels) to infer a Bayesian posterior over the underlying reward function dictating the user's preferences. We demonstrate the performance of the toolbox in simulation and present various applications of human-in-the-loop preference-based learning.", "date": "2022-12-21", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20221219-234115665", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20221219-234115665", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2208.04404", "resource_type": "monograph", "pub_year": "2022", "author_list": "Tucker, Maegan; Li, Kejun; et el." }, { "id": "https://authors.library.caltech.edu/records/ya1d9-y2y64", "eprint_id": 118462, "eprint_status": "archive", "datestamp": "2023-08-20 08:14:38", "lastmod": "2023-10-24 23:22:03", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Sun-Jennifer-J", "name": { "family": "Sun", "given": "Jennifer J." }, "orcid": "0000-0002-0906-6589" }, { "id": "Ulmer-Andrew", "name": { "family": "Ulmer", "given": "Andrew" } }, { "id": "Chakraborty-Dipam", "name": { "family": "Chakraborty", "given": "Dipam" } }, { "id": "Geuther-Brian", "name": { "family": "Geuther", "given": "Brian" }, "orcid": "0000-0002-7822-486X" }, { "id": "Hayes-Edward", "name": { "family": "Hayes", "given": "Edward" } }, { "id": "Jia-Heng", "name": { "family": "Jia", "given": "Heng" } }, { "id": "Kumar-Vivek", "name": { "family": "Kumar", "given": "Vivek" } }, { "id": "Partridge-Zachary", "name": { "family": "Partridge", "given": "Zachary" } }, { "id": "Robie-Alice-A", "name": { "family": "Robie", "given": "Alice" }, "orcid": "0000-0002-0784-2927" }, { "id": "Schretter-Catherine-E", "name": { "family": "Schretter", "given": "Catherine" }, "orcid": "0000-0002-3957-6838" }, { "id": "Sun-Chao", "name": { "family": "Sun", "given": "Chao" } }, { "id": "Sheppard-Keith", "name": { "family": "Sheppard", "given": "Keith" }, "orcid": "0000-0003-0842-9365" }, { "id": "Uttarwar-Param", "name": { "family": "Uttarwar", "given": "Param" } }, { "id": "Perona-P", "name": { "family": "Perona", "given": "Pietro" }, "orcid": "0000-0002-7583-5809" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Branson-Kristin", "name": { "family": "Branson", "given": "Kristin" }, "orcid": "0000-0002-5567-2512" }, { "id": "Kennedy-Ann", "name": { "family": "Kennedy", "given": "Ann" }, "orcid": "0000-0002-3782-0518" } ] }, "title": "The MABe22 Benchmarks for Representation Learning of Multi-Agent Behavior", "ispublished": "unpub", "full_text_status": "public", "note": "This work was generously supported by the Simons Collaboration on the Global Brain grant 543025 (to PP), NIH Award #R00MH117264 (to AK), NSF Award #1918839 (to YY), NSERC Award #PGSD3-532647-2019 (to JJS), as well as a gift from Charles and Lily Trimble (to PP). We would like to thank Tom Sproule for mouse breeding and dataset collection. The mouse dataset was supported by the National Institute of Health DA041668 (NIDA), DA048634 (NIDA, and Simons Foundation SFARI Director's Award) (to VK). We also greatly appreciate Google, Amazon, HHMI, and the Simons Foundation for sponsoring the MABe 2022 Challenge and Workshop.", "abstract": "Real-world behavior is often shaped by complex interactions between multiple agents. To scalably study multi-agent behavior, advances in unsupervised and self-supervised learning have enabled a variety of different behavioral representations to be learned from trajectory data. To date, there does not exist a unified set of benchmarks that can enable comparing methods quantitatively and systematically across a broad set of behavior analysis settings. We aim to address this by introducing a large-scale, multi-agent trajectory dataset from real-world behavioral neuroscience experiments that covers a range of behavior analysis tasks. Our dataset consists of trajectory data from common model organisms, with 9.6 million frames of mouse data and 4.4 million frames of fly data, in a variety of experimental settings, such as different strains, lengths of interaction, and optogenetic stimulation. A subset of the frames also consist of expert-annotated behavior labels. Improvements on our dataset corresponds to behavioral representations that work across multiple organisms and is able to capture differences for common behavior analysis tasks.", "date": "2022-12-21", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20221219-234042044", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20221219-234042044", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "Simons Foundation", "grant_number": "543025" }, { "agency": "NIH", "grant_number": "R00MH117264" }, { "agency": "NSF", "grant_number": "CCF-1918839" }, { "agency": "Natural Sciences and Engineering Research Council of Canada (NSERC)", "grant_number": "PGSD3-532647-2019" }, { "agency": "Charles and Lily Trimble" }, { "agency": "NIH", "grant_number": "DA041668" }, { "agency": "NIH", "grant_number": "DA048634" } ] }, "doi": "10.48550/arXiv.2207.10553", "resource_type": "monograph", "pub_year": "2022", "author_list": "Sun, Jennifer J.; Ulmer, Andrew; et el." }, { "id": "https://authors.library.caltech.edu/records/cwf24-dcd30", "eprint_id": 118408, "eprint_status": "archive", "datestamp": "2023-08-20 08:53:10", "lastmod": "2023-10-24 23:20:37", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Sun-Jennifer-J", "name": { "family": "Sun", "given": "Jennifer J." }, "orcid": "0000-0002-0906-6589" }, { "id": "Karashchuk-Pierre", "name": { "family": "Karashchuk", "given": "Pierre" }, "orcid": "0000-0001-6244-8239" }, { "id": "Dravid-Amil", "name": { "family": "Dravid", "given": "Amil" }, "orcid": "0000-0001-6007-0690" }, { "id": "Ryou-Serim", "name": { "family": "Ryou", "given": "Serim" }, "orcid": "0000-0003-1344-1158" }, { "id": "Fereidooni-Sonia", "name": { "family": "Fereidooni", "given": "Sonia" } }, { "id": "Tuthill-John-C", "name": { "family": "Tuthill", "given": "John C." }, "orcid": "0000-0002-5689-5806" }, { "id": "Katsaggelos-Aggelos", "name": { "family": "Katsaggelos", "given": "Aggelos" }, "orcid": "0000-0003-4554-0070" }, { "id": "Brunton-Bingni-W", "name": { "family": "Brunton", "given": "Bingni W." }, "orcid": "0000-0002-4831-3466" }, { "id": "Gkioxari-Georgia", "name": { "family": "Gkioxari", "given": "Georgia" } }, { "id": "Kennedy-Ann", "name": { "family": "Kennedy", "given": "Ann" }, "orcid": "0000-0002-3782-0518" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Perona-P", "name": { "family": "Perona", "given": "Pietro" }, "orcid": "0000-0002-7583-5809" } ] }, "title": "BKinD-3D: Self-Supervised 3D Keypoint Discovery from Multi-View Videos", "ispublished": "unpub", "full_text_status": "public", "note": "This work is generously supported by the Amazon AI4Science Fellowship (to JJS), NIH NINDS (R01NS102333 to JCT), and the Air Force Office of Scientific Research (AFOSR FA9550-19-1-0386 to BWB).\n\nSubmitted - 2212.07401.pdf
", "abstract": "Quantifying motion in 3D is important for studying the behavior of humans and other animals, but manual pose annotations are expensive and time-consuming to obtain. Self-supervised keypoint discovery is a promising strategy for estimating 3D poses without annotations. However, current keypoint discovery approaches commonly process single 2D views and do not operate in the 3D space. We propose a new method to perform self-supervised keypoint discovery in 3D from multi-view videos of behaving agents, without any keypoint or bounding box supervision in 2D or 3D. Our method uses an encoder-decoder architecture with a 3D volumetric heatmap, trained to reconstruct spatiotemporal differences across multiple views, in addition to joint length constraints on a learned 3D skeleton of the subject. In this way, we discover keypoints without requiring manual supervision in videos of humans and rats, demonstrating the potential of 3D keypoint discovery for studying behavior.", "date": "2022-12-20", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20221219-204745839", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20221219-204745839", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "Amazon AI4Science Fellowship" }, { "agency": "NIH", "grant_number": "R01NS102333" }, { "agency": "Air Force Office of Scientific Research (AFOSR)", "grant_number": "FA9550-19-1-0386" } ] }, "doi": "10.48550/arXiv.2212.07401", "primary_object": { "basename": "2212.07401.pdf", "url": "https://authors.library.caltech.edu/records/cwf24-dcd30/files/2212.07401.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Sun, Jennifer J.; Karashchuk, Pierre; et el." }, { "id": "https://authors.library.caltech.edu/records/m3zrt-p6h47", "eprint_id": 115568, "eprint_status": "archive", "datestamp": "2023-08-20 08:11:21", "lastmod": "2023-10-24 16:35:30", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Dorobantu-Victor-D", "name": { "family": "Dorobantu", "given": "Victor D." }, "orcid": "0000-0002-2797-7802" }, { "id": "Azizzadenesheli-Kamyar", "name": { "family": "Azizzadenesheli", "given": "Kamyar" }, "orcid": "0000-0001-8507-1868" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Compactly Restrictable Metric Policy Optimization Problems", "ispublished": "unpub", "full_text_status": "public", "keywords": "Continuous Markov Decision Processes, Reinforcement Learning, Optimal Control, Value Iteration, Selection Theorems, Sampled-Data, Physical Systems", "note": "Submitted May 15th, 2021. Resubmitted July 6th, 2022. This work was supported in part by DARPA and Beyond Limits. Victor D. Dorobantu was also supported in part by a Kortschak Fellowship.\n\nSubmitted - 2207.05850.pdf
", "abstract": "We study policy optimization problems for deterministic Markov decision processes (MDPs) with metric state and action spaces, which we refer to as Metric Policy Optimization Problems (MPOPs). Our goal is to establish theoretical results on the well-posedness of MPOPs that can characterize practically relevant continuous control systems. To do so, we define a special class of MPOPs called Compactly Restrictable MPOPs (CR-MPOPs), which are flexible enough to capture the complex behavior of robotic systems but specific enough to admit solutions using dynamic programming methods such as value iteration. We show how to arrive at CR-MPOPs using forward-invariance. We further show that our theoretical results on CR-MPOPs can be used to characterize feedback linearizable control affine systems.", "date": "2022-07-15", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20220714-212414777", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20220714-212414777", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "Defense Advanced Research Projects Agency (DARPA)" }, { "agency": "Beyond Limits" }, { "agency": "Kortschak Scholars Program" } ] }, "doi": "10.48550/arXiv.arXiv.2207.05850", "primary_object": { "basename": "2207.05850.pdf", "url": "https://authors.library.caltech.edu/records/m3zrt-p6h47/files/2207.05850.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Dorobantu, Victor D.; Azizzadenesheli, Kamyar; et el." }, { "id": "https://authors.library.caltech.edu/records/bme38-gm639", "eprint_id": 115570, "eprint_status": "archive", "datestamp": "2023-08-20 07:59:37", "lastmod": "2023-10-24 16:35:36", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Talukder-Sabera", "name": { "family": "Talukder", "given": "Sabera" } }, { "id": "Sun-Jennifer-J", "name": { "family": "Sun", "given": "Jennifer J." }, "orcid": "0000-0002-0906-6589" }, { "id": "Leonard-Matthew-K", "name": { "family": "Leonard", "given": "Matthew" } }, { "id": "Brunton-Bingni-W", "name": { "family": "Brunton", "given": "Bingni W." } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Deep Neural Imputation: A Framework for Recovering Incomplete Brain Recordings", "ispublished": "unpub", "full_text_status": "public", "note": "Attribution 4.0 International (CC BY 4.0) \n\nWe thank Albert Hao Li for thoughtful discussions and feedback throughout the project, Steve Peterson & Zoe Steine-Hanson for sharing their AJILE12 dataset knowledge, and Ann Kennedy for helpful conversations. This work was supported by an NSF Graduate Fellowship (to ST), NSERC Award #PGSD3-532647-2019 (to JJS), and the Moore Distinguished Scholar Program at Caltech (to BWB).\n\nSubmitted - 2206.08094.pdf
", "abstract": "Neuroscientists and neuroengineers have long relied on multielectrode neural recordings to study the brain. However, in a typical experiment, many factors corrupt neural recordings from individual electrodes, including electrical noise, movement artifacts, and faulty manufacturing. Currently, common practice is to discard these corrupted recordings, reducing already limited data that is difficult to collect. To address this challenge, we propose Deep Neural Imputation (DNI), a framework to recover missing values from electrodes by learning from data collected across spatial locations, days, and participants. We explore our framework with a linear nearest-neighbor approach and two deep generative autoencoders, demonstrating DNI's flexibility. One deep autoencoder models participants individually, while the other extends this architecture to model many participants jointly. We evaluate our models across 12 human participants implanted with multielectrode intracranial electrocorticography arrays; participants had no explicit task and behaved naturally across hundreds of recording hours. We show that DNI recovers not only time series but also frequency content, and further establish DNI's practical value by recovering significant performance on a scientifically-relevant downstream neural decoding task.", "date": "2022-07-15", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20220714-212423144", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20220714-212423144", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NSF Graduate Research Fellowship" }, { "agency": "Natural Sciences and Engineering Research Council of Canada (NSERC)", "grant_number": "PGSD3-532647-2019" }, { "agency": "Gordon and Betty Moore Foundation" } ] }, "doi": "10.48550/arXiv.arXiv.2206.08094", "primary_object": { "basename": "2206.08094.pdf", "url": "https://authors.library.caltech.edu/records/bme38-gm639/files/2206.08094.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Talukder, Sabera; Sun, Jennifer J.; et el." }, { "id": "https://authors.library.caltech.edu/records/5d294-ehk32", "eprint_id": 115569, "eprint_status": "archive", "datestamp": "2023-08-20 08:01:02", "lastmod": "2023-10-24 16:35:33", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Voloshin-Cameron", "name": { "family": "Voloshin", "given": "Cameron" } }, { "id": "Le-Hoang-M", "name": { "family": "Le", "given": "Hoang M." } }, { "id": "Chaudhuri-Swarat", "name": { "family": "Chaudhuri", "given": "Swarat" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Policy Optimization with Linear Temporal Logic Constraints", "ispublished": "unpub", "full_text_status": "public", "note": "Attribution 4.0 International (CC BY 4.0)\n\nSubmitted - 2206.09546.pdf
", "abstract": "We study the problem of policy optimization (PO) with linear temporal logic (LTL) constraints. The language of LTL allows flexible description of tasks that may be unnatural to encode as a scalar cost function. We consider LTL-constrained PO as a systematic framework, decoupling task specification from policy selection, and an alternative to the standard of cost shaping. With access to a generative model, we develop a model-based approach that enjoys a sample complexity analysis for guaranteeing both task satisfaction and cost optimality (through a reduction to a reachability problem). Empirically, our algorithm can achieve strong performance even in low sample regimes.", "date": "2022-07-15", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20220714-212419626", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20220714-212419626", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.arXiv.2206.09546", "primary_object": { "basename": "2206.09546.pdf", "url": "https://authors.library.caltech.edu/records/5d294-ehk32/files/2206.09546.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Voloshin, Cameron; Le, Hoang M.; et el." }, { "id": "https://authors.library.caltech.edu/records/v0rk2-qym53", "eprint_id": 114094, "eprint_status": "archive", "datestamp": "2023-08-20 07:20:30", "lastmod": "2023-10-23 23:21:40", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Taylor-Andrew-J", "name": { "family": "Taylor", "given": "Andrew J." }, "orcid": "0000-0002-5990-590X" }, { "id": "Dorobantu-Victor-D", "name": { "family": "Dorobantu", "given": "Victor D." }, "orcid": "0000-0002-2797-7802" }, { "id": "Cosner-Ryan-K", "name": { "family": "Cosner", "given": "Ryan K." }, "orcid": "0000-0002-4035-1425" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Ames-A-D", "name": { "family": "Ames", "given": "Aaron D." }, "orcid": "0000-0003-0848-3177" } ] }, "title": "Safety of Sampled-Data Systems with Control Barrier Functions via Approximate Discrete Time Models", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 2203.11470.pdf
", "abstract": "Control Barrier Functions (CBFs) have been demonstrated to be a powerful tool for safety-critical controller design for nonlinear systems. Existing design paradigms do not address the gap between theory (controller design with continuous time models) and practice (the discrete time sampled implementation of the resulting controllers); this can lead to poor performance and violations of safety for hardware instantiations. We propose an approach to close this gap by synthesizing sampled-data counterparts to these CBF-based controllers using approximate discrete time models and Sampled-Data Control Barrier Functions (SD-CBFs). Using properties of a system's continuous time model, we establish a relationship between SD-CBFs and a notion of practical safety for sampled-data systems. Furthermore, we construct convex optimization-based controllers that formally endow nonlinear systems with safety guarantees in practice. We demonstrate the efficacy of these controllers in simulation.", "date": "2022-03-28", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20220325-224027516", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20220325-224027516", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2203.11470", "primary_object": { "basename": "2203.11470.pdf", "url": "https://authors.library.caltech.edu/records/v0rk2-qym53/files/2203.11470.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Taylor, Andrew J.; Dorobantu, Victor D.; et el." }, { "id": "https://authors.library.caltech.edu/records/atx9r-ja815", "eprint_id": 114092, "eprint_status": "archive", "datestamp": "2023-08-20 07:11:36", "lastmod": "2023-10-23 23:21:38", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Cosner-Ryan-K", "name": { "family": "Cosner", "given": "Ryan K." }, "orcid": "0000-0002-4035-1425" }, { "id": "Jimenez-Rodriguez-Ivan-D", "name": { "family": "Jimenez Rodriguez", "given": "Ivan D." } }, { "id": "Moln\u00e1r-Tam\u00e1s-G", "name": { "family": "Molnar", "given": "Tamas G." }, "orcid": "0000-0002-9379-7121" }, { "id": "Ubellacker-Wyatt-L", "name": { "family": "Ubellacker", "given": "Wyatt" }, "orcid": "0000-0002-4732-6185" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Ames-A-D", "name": { "family": "Ames", "given": "Aaron D." }, "orcid": "0000-0003-0848-3177" }, { "id": "Bouman-K-L", "name": { "family": "Bouman", "given": "Katherine L." }, "orcid": "0000-0003-0077-4367" } ] }, "title": "Self-Supervised Online Learning for Safety-Critical Control using Stereo Vision", "ispublished": "unpub", "full_text_status": "public", "note": "Attribution 4.0 International (CC BY 4.0).\n\nThis research is supported in part by the National Science Foundation CPS Award #1932091, Dow (#227027AT), BP p.l.c., AeroVironment.\n\nSubmitted - 2203.01404.pdf
", "abstract": "With the increasing prevalence of complex vision-based sensing methods for use in obstacle identification and state estimation, characterizing environment-dependent measurement errors has become a difficult and essential part of modern robotics. This paper presents a self-supervised learning approach to safety-critical control. In particular, the uncertainty associated with stereo vision is estimated, and adapted online to new visual environments, wherein this estimate is leveraged in a safety-critical controller in a robust fashion. To this end, we propose an algorithm that exploits the structure of stereo-vision to learn an uncertainty estimate without the need for ground-truth data. We then robustify existing Control Barrier Function-based controllers to provide safety in the presence of this uncertainty estimate. We demonstrate the efficacy of our method on a quadrupedal robot in a variety of environments. When not using our method safety is violated. With offline training alone we observe the robot is safe, but overly-conservative. With our online method the quadruped remains safe and conservatism is reduced.", "date": "2022-03-25", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20220325-220806703", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20220325-220806703", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NSF", "grant_number": "CNS-1932091" }, { "agency": "Dow Chemical Company", "grant_number": "227027AT" }, { "agency": "BP" }, { "agency": "AeroVironment" } ] }, "doi": "10.48550/arXiv.2203.01404", "primary_object": { "basename": "2203.01404.pdf", "url": "https://authors.library.caltech.edu/records/atx9r-ja815/files/2203.01404.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Cosner, Ryan K.; Jimenez Rodriguez, Ivan D.; et el." }, { "id": "https://authors.library.caltech.edu/records/72a5j-ate19", "eprint_id": 113578, "eprint_status": "archive", "datestamp": "2023-08-20 04:22:16", "lastmod": "2023-10-23 23:07:19", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Zhan-Eric", "name": { "family": "Zhan", "given": "Eric" } }, { "id": "Sun-Jennifer-J", "name": { "family": "Sun", "given": "Jennifer J." }, "orcid": "0000-0002-0906-6589" }, { "id": "Kennedy-Ann", "name": { "family": "Kennedy", "given": "Ann" }, "orcid": "0000-0002-3782-0518" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Chaudhuri-Swarat", "name": { "family": "Chaudhuri", "given": "Swarat" } } ] }, "title": "Unsupervised Learning of Neurosymbolic Encoders", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 2107.13132.pdf
", "abstract": "We present a framework for the unsupervised learning of neurosymbolic encoders, i.e., encoders obtained by composing neural networks with symbolic programs from a domain-specific language. Such a framework can naturally incorporate symbolic expert knowledge into the learning process and lead to more interpretable and factorized latent representations than fully neural encoders. Also, models learned this way can have downstream impact, as many analysis workflows can benefit from having clean programmatic descriptions. We ground our learning algorithm in the variational autoencoding (VAE) framework, where we aim to learn a neurosymbolic encoder in conjunction with a standard decoder. Our algorithm integrates standard VAE-style training with modern program synthesis techniques. We evaluate our method on learning latent representations for real-world trajectory data from animal biology and sports analytics. We show that our approach offers significantly better separation than standard VAEs and leads to practical gains on downstream tasks.", "date": "2022-03-01", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20220224-200805115", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20220224-200805115", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2107.13132", "primary_object": { "basename": "2107.13132.pdf", "url": "https://authors.library.caltech.edu/records/72a5j-ate19/files/2107.13132.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Zhan, Eric; Sun, Jennifer J.; et el." }, { "id": "https://authors.library.caltech.edu/records/mwf88-ytc90", "eprint_id": 113585, "eprint_status": "archive", "datestamp": "2023-08-20 06:04:01", "lastmod": "2023-10-23 23:07:32", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Tseng-Albert", "name": { "family": "Tseng", "given": "Albert" } }, { "id": "Sun-Jennifer-J", "name": { "family": "Sun", "given": "Jennifer J." }, "orcid": "0000-0002-0906-6589" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Automatic Synthesis of Diverse Weak Supervision Sources for Behavior Analysis", "ispublished": "unpub", "full_text_status": "public", "note": "We thank Adith Swaminathan of Microsoft Research and Pietro Perona of Caltech for their invaluable feedback and helpful discussions regarding this work. We also thank Microsoft Research for the compute resources for our experiments. This work is partially supported by NSF Award #1918839 (YY) and NSERC Award #PGSD3-532647-2019 (JJS).\n\nSubmitted - 2111.15186.pdf
", "abstract": "Obtaining annotations for large training sets is expensive, especially in behavior analysis settings where domain knowledge is required for accurate annotations. Weak supervision has been studied to reduce annotation costs by using weak labels from task-level labeling functions to augment ground truth labels. However, domain experts are still needed to hand-craft labeling functions for every studied task. To reduce expert effort, we present AutoSWAP: a framework for automatically synthesizing data-efficient task-level labeling functions. The key to our approach is to efficiently represent expert knowledge in a reusable domain specific language and domain-level labeling functions, with which we use state-of-the-art program synthesis techniques and a small labeled dataset to generate labeling functions. Additionally, we propose a novel structural diversity cost that allows for direct synthesis of diverse sets of labeling functions with minimal overhead, further improving labeling function data efficiency. We evaluate AutoSWAP in three behavior analysis domains and demonstrate that AutoSWAP outperforms existing approaches using only a fraction of the data. Our results suggest that AutoSWAP is an effective way to automatically generate labeling functions that can significantly reduce expert effort for behavior analysis.", "date": "2022-02-28", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20220224-200830238", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20220224-200830238", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NSF", "grant_number": "CCF-1918839" }, { "agency": "Natural Sciences and Engineering Research Council of Canada (NSERC)", "grant_number": "PGSD3-532647-2019" } ] }, "doi": "10.48550/arXiv.2111.15186", "primary_object": { "basename": "2111.15186.pdf", "url": "https://authors.library.caltech.edu/records/mwf88-ytc90/files/2111.15186.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Tseng, Albert; Sun, Jennifer J.; et el." }, { "id": "https://authors.library.caltech.edu/records/c5hxv-yxh35", "eprint_id": 113583, "eprint_status": "archive", "datestamp": "2023-08-20 05:33:27", "lastmod": "2023-10-23 23:07:30", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Bernstein-Jeremy-D", "name": { "family": "Bernstein", "given": "Jeremy" }, "orcid": "0000-0001-9110-7476" }, { "id": "Farhang-Alex", "name": { "family": "Farhang", "given": "Alex" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Kernel Interpolation as a Bayes Point Machine", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 2110.04274.pdf
", "abstract": "A Bayes point machine is a single classifier that approximates the majority decision of an ensemble of classifiers. This paper observes that kernel interpolation is a Bayes point machine for Gaussian process classification. This observation facilitates the transfer of results from both ensemble theory as well as an area of convex geometry known as Brunn-Minkowski theory to derive PAC-Bayes risk bounds for kernel interpolation. Since large margin, infinite width neural networks are kernel interpolators, the paper's findings may help to explain generalisation in neural networks more broadly. Supporting this idea, the paper finds evidence that large margin, finite width neural networks behave like Bayes point machines too.", "date": "2022-02-28", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20220224-200822492", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20220224-200822492", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2110.04274", "primary_object": { "basename": "2110.04274.pdf", "url": "https://authors.library.caltech.edu/records/c5hxv-yxh35/files/2110.04274.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Bernstein, Jeremy; Farhang, Alex; et el." }, { "id": "https://authors.library.caltech.edu/records/2j5rs-cvh78", "eprint_id": 113589, "eprint_status": "archive", "datestamp": "2023-08-20 06:19:06", "lastmod": "2023-10-23 23:07:43", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Cosner-Ryan-K", "name": { "family": "Cosner", "given": "Ryan K." }, "orcid": "0000-0002-4035-1425" }, { "id": "Tucker-Maegan", "name": { "family": "Tucker", "given": "Maegan" }, "orcid": "0000-0001-7363-6809" }, { "id": "Taylor-Andrew-J", "name": { "family": "Taylor", "given": "Andrew J." }, "orcid": "0000-0002-5990-590X" }, { "id": "Li-Kejun", "name": { "family": "Li", "given": "Kejun" } }, { "id": "Moln\u00e1r-Tam\u00e1s-G", "name": { "family": "Moln\u00e1r", "given": "Tam\u00e1s G." }, "orcid": "0000-0002-9379-7121" }, { "id": "Ubellacker-Wyatt-L", "name": { "family": "Ubellacker", "given": "Wyatt" }, "orcid": "0000-0002-4732-6185" }, { "id": "Alan-Anil", "name": { "family": "Alan", "given": "Anil" }, "orcid": "0000-0002-9778-8249" }, { "id": "Orosz-G\u00e1bor", "name": { "family": "Orosz", "given": "G\u00e1bor" }, "orcid": "0000-0002-9000-3736" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Ames-A-D", "name": { "family": "Ames", "given": "Aaron D." }, "orcid": "0000-0003-0848-3177" } ] }, "title": "Safety-Aware Preference-Based Learning for Safety-Critical Control", "ispublished": "unpub", "full_text_status": "public", "keywords": "Preference-Based Learning, Control Barrier Functions, Safety-Critical Control, Robotics", "note": "\u00a9 2022 R.K. Cosner, M. Tucker, A.J. Taylor, K. Li, T.G. Molnar, W. Ubellacker, A. Alan, G. Orosz, Y. Yue & A.D. Ames. Attribution 4.0 International (CC BY 4.0).\n\nSubmitted - 2112.08516.pdf
", "abstract": "Bringing dynamic robots into the wild requires a tenuous balance between performance and safety. Yet controllers designed to provide robust safety guarantees often result in conservative behavior, and tuning these controllers to find the ideal trade-off between performance and safety typically requires domain expertise or a carefully constructed reward function. This work presents a design paradigm for systematically achieving behaviors that balance performance and robust safety by integrating safety-aware Preference-Based Learning (PBL) with Control Barrier Functions (CBFs). Fusing these concepts -- safety-aware learning and safety-critical control -- gives a robust means to achieve safe behaviors on complex robotic systems in practice. We demonstrate the capability of this design paradigm to achieve safe and performant perception-based autonomous operation of a quadrupedal robot both in simulation and experimentally on hardware.", "date": "2022-02-28", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20220224-200843937", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20220224-200843937", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2112.08516", "primary_object": { "basename": "2112.08516.pdf", "url": "https://authors.library.caltech.edu/records/2j5rs-cvh78/files/2112.08516.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Cosner, Ryan K.; Tucker, Maegan; et el." }, { "id": "https://authors.library.caltech.edu/records/j33yw-5g106", "eprint_id": 113586, "eprint_status": "archive", "datestamp": "2023-08-20 06:13:42", "lastmod": "2023-12-22 23:41:44", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Sun-Jennifer-J", "name": { "family": "Sun", "given": "Jennifer J." }, "orcid": "0000-0002-0906-6589" }, { "id": "Ryou-Serim", "name": { "family": "Ryou", "given": "Serim" } }, { "id": "Goldshmid-Roni-H", "name": { "family": "Goldshmid", "given": "Roni" }, "orcid": "0000-0001-9095-3259" }, { "id": "Weissbourd-Brandon", "name": { "family": "Weissbourd", "given": "Brandon" }, "orcid": "0000-0001-5422-3873" }, { "id": "Dabiri-J-O", "name": { "family": "Dabiri", "given": "John" }, "orcid": "0000-0002-6722-9008" }, { "id": "Anderson-D-J", "name": { "family": "Anderson", "given": "David J." }, "orcid": "0000-0001-6175-3872" }, { "id": "Kennedy-Ann", "name": { "family": "Kennedy", "given": "Ann" }, "orcid": "0000-0002-3782-0518" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Perona-P", "name": { "family": "Perona", "given": "Pietro" }, "orcid": "0000-0002-7583-5809" } ] }, "title": "Self-Supervised Keypoint Discovery in Behavioral Videos", "ispublished": "unpub", "full_text_status": "public", "note": "This work was generously supported by the Simons Collaboration on the Global Brain grant 543025 (to PP and DJA), NIH Award #R00MH117264 (to AK), NSF Award #1918839 (to YY), NINDS Award #K99NS119749 (to BW), NIH Award #R01MH123612 (to DJA, PP, and SR), NSERC Award #PGSD3-532647-2019 (to JJS), as well as a gift from Charles and Lily Trimble (to PP).\n\nSubmitted - 2112.05121.pdf
", "abstract": "We propose a method for learning the posture and structure of agents from unlabelled behavioral videos. Starting from the observation that behaving agents are generally the main sources of movement in behavioral videos, our method uses an encoder-decoder architecture with a geometric bottleneck to reconstruct the difference between video frames. By focusing only on regions of movement, our approach works directly on input videos without requiring manual annotations, such as keypoints or bounding boxes. Experiments on a variety of agent types (mouse, fly, human, jellyfish, and trees) demonstrate the generality of our approach and reveal that our discovered keypoints represent semantically meaningful body parts, which achieve state-of-the-art performance on keypoint regression among self-supervised methods. Additionally, our discovered keypoints achieve comparable performance to supervised keypoints on downstream tasks, such as behavior classification, suggesting that our method can dramatically reduce the cost of model training vis-a-vis supervised methods.", "date": "2022-02-28", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20220224-200833645", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20220224-200833645", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "Simons Foundation", "grant_number": "543025" }, { "agency": "NIH", "grant_number": "R00MH117264" }, { "agency": "NSF", "grant_number": "CCF-1918839" }, { "agency": "NIH", "grant_number": "K99NS119749" }, { "agency": "NIH", "grant_number": "R01MH123612" }, { "agency": "Natural Sciences and Engineering Research Council of Canada (NSERC)", "grant_number": "PGSD3-532647-2019" }, { "agency": "Charles and Lily Trimble" } ] }, "local_group": { "items": [ { "id": "Tianqiao-and-Chrissy-Chen-Institute-for-Neuroscience" }, { "id": "GALCIT" }, { "id": "Division-of-Biology-and-Biological-Engineering" } ] }, "doi": "10.48550/arXiv.2112.05121", "primary_object": { "basename": "2112.05121.pdf", "url": "https://authors.library.caltech.edu/records/j33yw-5g106/files/2112.05121.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Sun, Jennifer J.; Ryou, Serim; et el." }, { "id": "https://authors.library.caltech.edu/records/pc93q-chx28", "eprint_id": 113576, "eprint_status": "archive", "datestamp": "2023-08-20 03:40:11", "lastmod": "2023-10-23 23:07:14", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Tjandrasuwita-Megan", "name": { "family": "Tjandrasuwita", "given": "Megan" } }, { "id": "Sun-Jennifer-J", "name": { "family": "Sun", "given": "Jennifer J." }, "orcid": "0000-0002-0906-6589" }, { "id": "Kennedy-Ann", "name": { "family": "Kennedy", "given": "Ann" }, "orcid": "0000-0002-3782-0518" }, { "id": "Chaudhuri-Swarat", "name": { "family": "Chaudhuri", "given": "Swarat" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Interpreting Expert Annotation Differences in Animal Behavior", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 2106.06114.pdf
", "abstract": "Hand-annotated data can vary due to factors such as subjective differences, intra-rater variability, and differing annotator expertise. We study annotations from different experts who labelled the same behavior classes on a set of animal behavior videos, and observe a variation in annotation styles. We propose a new method using program synthesis to help interpret annotation differences for behavior analysis. Our model selects relevant trajectory features and learns a temporal filter as part of a program, which corresponds to estimated importance an annotator places on that feature at each timestamp. Our experiments on a dataset from behavioral neuroscience demonstrate that compared to baseline approaches, our method is more accurate at capturing annotator labels and learns interpretable temporal filters. We believe that our method can lead to greater reproducibility of behavior annotations used in scientific studies. We plan to release our code.", "date": "2022-02-25", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20220224-200758198", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20220224-200758198", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2106.06114", "primary_object": { "basename": "2106.06114.pdf", "url": "https://authors.library.caltech.edu/records/pc93q-chx28/files/2106.06114.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Tjandrasuwita, Megan; Sun, Jennifer J.; et el." }, { "id": "https://authors.library.caltech.edu/records/ke5cf-9cj15", "eprint_id": 113606, "eprint_status": "archive", "datestamp": "2023-08-20 06:58:33", "lastmod": "2023-10-23 23:08:11", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Jimenez-Rodriguez-Ivan-Dario", "name": { "family": "Jimenez Rodriguez", "given": "Ivan Dario" }, "orcid": "0000-0001-9065-5227" }, { "id": "Ames-A-D", "name": { "family": "Ames", "given": "Aaron D." }, "orcid": "0000-0003-0848-3177" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "LyaNet: A Lyapunov Framework for Training Neural ODEs", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 2202.02526.pdf
", "abstract": "We propose a method for training ordinary differential equations by using a control-theoretic Lyapunov condition for stability. Our approach, called LyaNet, is based on a novel Lyapunov loss formulation that encourages the inference dynamics to converge quickly to the correct prediction. Theoretically, we show that minimizing Lyapunov loss guarantees exponential convergence to the correct solution and enables a novel robustness guarantee. We also provide practical algorithms, including one that avoids the cost of backpropagating through a solver or using the adjoint method. Relative to standard Neural ODE training, we empirically find that LyaNet can offer improved prediction performance, faster convergence of inference dynamics, and improved adversarial robustness. Our code available at https://github.com/ivandariojr/LyapunovLearning.", "date": "2022-02-25", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20220224-200943137", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20220224-200943137", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2202.02526", "primary_object": { "basename": "2202.02526.pdf", "url": "https://authors.library.caltech.edu/records/ke5cf-9cj15/files/2202.02526.pdf" }, "resource_type": "monograph", "pub_year": "2022", "author_list": "Jimenez Rodriguez, Ivan Dario; Ames, Aaron D.; et el." }, { "id": "https://authors.library.caltech.edu/records/enkxv-30g64", "eprint_id": 109916, "eprint_status": "archive", "datestamp": "2023-08-20 03:39:14", "lastmod": "2023-10-23 18:12:48", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Ferber-Aaron", "name": { "family": "Ferber", "given": "Aaron" } }, { "id": "Song-Jialin", "name": { "family": "Song", "given": "Jialin" } }, { "id": "Dilkina-Bistra", "name": { "family": "Dilkina", "given": "Bistra" }, "orcid": "0000-0002-6784-473X" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Learning Pseudo-Backdoors for Mixed Integer Programs", "ispublished": "unpub", "full_text_status": "public", "note": "\u00a9 2021, Association for the Advancement of Artificial Intelligence.\n\nSubmitted - 2106.05080.pdf
", "abstract": "We propose a machine learning approach for quickly solving Mixed Integer Programs (MIP) by learning to prioritize a set of decision variables, which we call pseudo-backdoors, for branching that results in faster solution times. Learning-based approaches have seen success in the area of solving combinatorial optimization problems by being able to flexibly leverage common structures in a given distribution of problems. Our approach takes inspiration from the concept of strong backdoors, which corresponds to a small set of variables such that only branching on these variables yields an optimal integral solution and a proof of optimality. Our notion of pseudo-backdoors corresponds to a small set of variables such that only branching on them leads to faster solve time (which can be solver dependent). A key advantage of pseudo-backdoors over strong backdoors is that they are much amenable to data-driven identification or prediction. Our proposed method learns to estimate the solver performance of a proposed pseudo-backdoor, using a labeled dataset collected on a set of training MIP instances. This model can then be used to identify high-quality pseudo-backdoors on new MIP instances from the same distribution. We evaluate our method on the generalized independent set problems and find that our approach can efficiently identify high-quality pseudo-backdoors. In addition, we compare our learned approach against Gurobi, a state-of-the-art MIP solver, demonstrating that our method can be used to improve solver performance.", "date": "2021-07-19", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20210719-210128990", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20210719-210128990", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2106.05080", "primary_object": { "basename": "2106.05080.pdf", "url": "https://authors.library.caltech.edu/records/enkxv-30g64/files/2106.05080.pdf" }, "resource_type": "monograph", "pub_year": "2021", "author_list": "Ferber, Aaron; Song, Jialin; et el." }, { "id": "https://authors.library.caltech.edu/records/qcp8d-87r65", "eprint_id": 109396, "eprint_status": "archive", "datestamp": "2023-08-20 03:14:29", "lastmod": "2023-10-23 17:54:50", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Yin-Tianwei", "name": { "family": "Yin", "given": "Tianwei" } }, { "id": "Wu-Zihui", "name": { "family": "Wu", "given": "Zihui" } }, { "id": "Sun-He", "name": { "family": "Sun", "given": "He" }, "orcid": "0000-0003-1526-6787" }, { "id": "Dalca-Adrian-V", "name": { "family": "Dalca", "given": "Adrian V." }, "orcid": "0000-0002-8422-0136" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Bouman-K-L", "name": { "family": "Bouman", "given": "Katherine L." }, "orcid": "0000-0003-0077-4367" } ] }, "title": "End-to-End Sequential Sampling and Reconstruction for MR Imaging", "ispublished": "unpub", "full_text_status": "public", "note": "Code and supplementary materials are available at this http URL http://imaging.cms.caltech.edu/seq-mri\n\nSubmitted - 2105.06460.pdf
", "abstract": "Accelerated MRI shortens acquisition time by subsampling in the measurement k-space. Recovering a high-fidelity anatomical image from subsampled measurements requires close cooperation between two components: (1) a sampler that chooses the subsampling pattern and (2) a reconstructor that recovers images from incomplete measurements. In this paper, we leverage the sequential nature of MRI measurements, and propose a fully differentiable framework that jointly learns a sequential sampling policy simultaneously with a reconstruction strategy. This co-designed framework is able to adapt during acquisition in order to capture the most informative measurements for a particular target (Figure 1). Experimental results on the fastMRI knee dataset demonstrate that the proposed approach successfully utilizes intermediate information during the sampling process to boost reconstruction performance. In particular, our proposed method outperforms the current state-of-the-art learned k-space sampling baseline on up to 96.96% of test samples. We also investigate the individual and collective benefits of the sequential sampling and co-design strategies. Code and more visualizations are available at this http URL [http://imaging.cms.caltech.edu/seq-mri]", "date": "2021-06-07", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20210604-142545306", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20210604-142545306", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "local_group": { "items": [ { "id": "Astronomy-Department" } ] }, "doi": "10.48550/arXiv.2105.06460", "primary_object": { "basename": "2105.06460.pdf", "url": "https://authors.library.caltech.edu/records/qcp8d-87r65/files/2105.06460.pdf" }, "resource_type": "monograph", "pub_year": "2021", "author_list": "Yin, Tianwei; Wu, Zihui; et el." }, { "id": "https://authors.library.caltech.edu/records/yej1d-d2j82", "eprint_id": 109027, "eprint_status": "archive", "datestamp": "2023-08-20 02:39:55", "lastmod": "2023-12-22 23:38:33", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Sun-Jennifer-J", "name": { "family": "Sun", "given": "Jennifer J." }, "orcid": "0000-0002-0906-6589" }, { "id": "Karigo-Tomomi", "name": { "family": "Karigo", "given": "Tomomi" } }, { "id": "Chakraborty-Dipam", "name": { "family": "Chakraborty", "given": "Dipam" } }, { "id": "Mohanty-Sharada-P", "name": { "family": "Mohanty", "given": "Sharada P." } }, { "id": "Anderson-D-J", "name": { "family": "Anderson", "given": "David J." }, "orcid": "0000-0001-6175-3872" }, { "id": "Perona-P", "name": { "family": "Perona", "given": "Pietro" }, "orcid": "0000-0002-7583-5809" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Kennedy-A", "name": { "family": "Kennedy", "given": "Ann" }, "orcid": "0000-0002-3782-0518" } ] }, "title": "The Multi-Agent Behavior Dataset: Mouse Dyadic Social Interactions", "ispublished": "unpub", "full_text_status": "public", "note": "We would like to thank the researchers at the David Anderson\nResearch Group at Caltech for this collaboration and the recording and annotation of the mouse behavior datasets. We are grateful to the team at AICrowd for the support and hosting our dataset challenge, as well as Northwestern University and Amazon Sagemaker for funding our challenge prizes. This work was generously supported by the Simons Collaboration on the Global Brain grant 543025 (to PP), NIH Award #K99MH117264 (to AK), NSF Award #1918839 (to YY), and NSERC Award #PGSD3-532647-2019 (to JJS).\n\nSubmitted - 2104.02710.pdf
", "abstract": "Multi-agent behavior modeling aims to understand the interactions that occur between agents. We present a multi-agent dataset from behavioral neuroscience, the Caltech Mouse Social Interactions (CalMS21) Dataset. Our dataset consists of trajectory data of social interactions, recorded from videos of freely behaving mice in a standard resident-intruder assay. The CalMS21 dataset is part of the Multi-Agent Behavior Challenge 2021 and for our next step, our goal is to incorporate datasets from other domains studying multi-agent behavior. \n\nTo help accelerate behavioral studies, the CalMS21 dataset provides a benchmark to evaluate the performance of automated behavior classification methods in three settings: (1) for training on large behavioral datasets all annotated by a single annotator, (2) for style transfer to learn inter-annotator differences in behavior definitions, and (3) for learning of new behaviors of interest given limited training data. The dataset consists of 6 million frames of unlabelled tracked poses of interacting mice, as well as over 1 million frames with tracked poses and corresponding frame-level behavior annotations. The challenge of our dataset is to be able to classify behaviors accurately using both labelled and unlabelled tracking data, as well as being able to generalize to new annotators and behaviors.", "date": "2021-05-10", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20210510-093610124", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20210510-093610124", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "Simons Foundation", "grant_number": "543025" }, { "agency": "NIH", "grant_number": "K99MH117264" }, { "agency": "NSF", "grant_number": "IIS-1918839" }, { "agency": "Natural Sciences and Engineering Research Council of Canada (NSERC)", "grant_number": "PGSD3-532647-2019" } ] }, "local_group": { "items": [ { "id": "Tianqiao-and-Chrissy-Chen-Institute-for-Neuroscience" }, { "id": "Division-of-Biology-and-Biological-Engineering" } ] }, "doi": "10.48550/arXiv.2104.02710", "primary_object": { "basename": "2104.02710.pdf", "url": "https://authors.library.caltech.edu/records/yej1d-d2j82/files/2104.02710.pdf" }, "resource_type": "monograph", "pub_year": "2021", "author_list": "Sun, Jennifer J.; Karigo, Tomomi; et el." }, { "id": "https://authors.library.caltech.edu/records/4g66e-dhs76", "eprint_id": 108309, "eprint_status": "archive", "datestamp": "2023-08-20 02:08:59", "lastmod": "2023-10-23 16:55:56", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Bernstein-Jeremy-D", "name": { "family": "Bernstein", "given": "Jeremy" }, "orcid": "0000-0001-9110-7476" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Computing the Information Content of Trained Neural Networks", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 2103.01045.pdf
", "abstract": "How much information does a learning algorithm extract from the training data and store in a neural network's weights? Too much, and the network would overfit to the training data. Too little, and the network would not fit to anything at all. Na\u00efvely, the amount of information the network stores should scale in proportion to the number of trainable weights. This raises the question: how can neural networks with vastly more weights than training data still generalise? A simple resolution to this conundrum is that the number of weights is usually a bad proxy for the actual amount of information stored. For instance, typical weight vectors may be highly compressible. Then another question occurs: is it possible to compute the actual amount of information stored? This paper derives both a consistent estimator and a closed-form upper bound on the information content of infinitely wide neural networks. The derivation is based on an identification between neural information content and the negative log probability of a Gaussian orthant. This identification yields bounds that analytically control the generalisation behaviour of the entire solution space of infinitely wide networks. The bounds have a simple dependence on both the network architecture and the training data. Corroborating the findings of Valle-P\u00e9rez et al. (2019), who conducted a similar analysis using approximate Gaussian integration techniques, the bounds are found to be both non-vacuous and correlated with the empirical generalisation behaviour at finite width.", "date": "2021-03-04", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20210304-095340677", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20210304-095340677", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2103.01045", "primary_object": { "basename": "2103.01045.pdf", "url": "https://authors.library.caltech.edu/records/4g66e-dhs76/files/2103.01045.pdf" }, "resource_type": "monograph", "pub_year": "2021", "author_list": "Bernstein, Jeremy and Yue, Yisong" }, { "id": "https://authors.library.caltech.edu/records/b6a7k-k8085", "eprint_id": 108203, "eprint_status": "archive", "datestamp": "2023-08-20 01:26:18", "lastmod": "2023-10-23 16:32:06", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Liu-Anqi", "name": { "family": "Liu", "given": "Anqi" } }, { "id": "Liu-Hao", "name": { "family": "Liu", "given": "Hao" }, "orcid": "0000-0002-7405-1578" }, { "id": "Li-Tongxin", "name": { "family": "Li", "given": "Tongxin" }, "orcid": "0000-0002-9806-8964" }, { "id": "Karimi-Bidhendi-Saeed", "name": { "family": "Karimi-Bidhendi", "given": "Saeed" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Anandkumar-A", "name": { "family": "Anandkumar", "given": "Anima" } } ] }, "title": "Disentangling Observed Causal Effects from Latent Confounders using Method of Moments", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 2101.06614.pdf
", "abstract": "Discovering the complete set of causal relations among a group of variables is a challenging unsupervised learning problem. Often, this challenge is compounded by the fact that there are latent or hidden confounders. When only observational data is available, the problem is ill-posed, i.e. the causal relationships are non-identifiable unless strong modeling assumptions are made. When interventions are available, we provide guarantees on identifiability and learnability under mild assumptions. We assume a linear structural equation model (SEM) with independent latent factors and directed acyclic graph (DAG) relationships among the observables. Since the latent variable inference is based on independent component analysis (ICA), we call this model SEM-ICA. We use the method of moments principle to establish model identifiability. We develop efficient algorithms based on coupled tensor decomposition with linear constraints to obtain scalable and guaranteed solutions. Thus, we provide a principled approach to tackling the joint problem of causal discovery and latent variable inference.", "date": "2021-02-26", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20210225-132714927", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20210225-132714927", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2101.06614", "primary_object": { "basename": "2101.06614.pdf", "url": "https://authors.library.caltech.edu/records/b6a7k-k8085/files/2101.06614.pdf" }, "resource_type": "monograph", "pub_year": "2021", "author_list": "Liu, Anqi; Liu, Hao; et el." }, { "id": "https://authors.library.caltech.edu/records/1bqjb-6nf06", "eprint_id": 107568, "eprint_status": "archive", "datestamp": "2023-08-20 00:17:23", "lastmod": "2023-10-23 16:01:25", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Talukder-Sabera", "name": { "family": "Talukder", "given": "Sabera" } }, { "id": "Raghavan-Guruprasad", "name": { "family": "Raghavan", "given": "Guruprasad" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Architecture Agnostic Neural Networks", "ispublished": "unpub", "full_text_status": "public", "note": "Attribution 4.0 International (CC BY 4.0).\n\nSubmitted - 2011.02712.pdf
", "abstract": "In this paper, we explore an alternate method for synthesizing neural network architectures, inspired by the brain's stochastic synaptic pruning. During a person's lifetime, numerous distinct neuronal architectures are responsible for performing the same tasks. This indicates that biological neural networks are, to some degree, architecture agnostic. However, artificial networks rely on their fine-tuned weights and hand-crafted architectures for their remarkable performance. This contrast begs the question: Can we build artificial architecture agnostic neural networks? To ground this study we utilize sparse, binary neural networks that parallel the brain's circuits. Within this sparse, binary paradigm we sample many binary architectures to create families of architecture agnostic neural networks not trained via backpropagation. These high-performing network families share the same sparsity, distribution of binary weights, and succeed in both static and dynamic tasks. In summation, we create an architecture manifold search procedure to discover families or architecture agnostic neural networks.", "date": "2021-01-20", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20210119-161636048", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20210119-161636048", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2011.02712", "primary_object": { "basename": "2011.02712.pdf", "url": "https://authors.library.caltech.edu/records/1bqjb-6nf06/files/2011.02712.pdf" }, "resource_type": "monograph", "pub_year": "2021", "author_list": "Talukder, Sabera; Raghavan, Guruprasad; et el." }, { "id": "https://authors.library.caltech.edu/records/q6qth-h9r71", "eprint_id": 107564, "eprint_status": "archive", "datestamp": "2023-08-20 00:22:25", "lastmod": "2023-10-23 16:01:15", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Barnum-George", "name": { "family": "Barnum", "given": "George" } }, { "id": "Talukder-Sabera", "name": { "family": "Talukder", "given": "Sabera" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "On the Benefits of Early Fusion in Multimodal Representation Learning", "ispublished": "unpub", "full_text_status": "public", "note": "Attribution 4.0 International (CC BY 4.0).\n\nSubmitted - 2011.07191.pdf
", "abstract": "Intelligently reasoning about the world often requires integrating data from multiple modalities, as any individual modality may contain unreliable or incomplete information. Prior work in multimodal learning fuses input modalities only after significant independent processing. On the other hand, the brain performs multimodal processing almost immediately. This divide between conventional multimodal learning and neuroscience suggests that a detailed study of early multimodal fusion could improve artificial multimodal representations. To facilitate the study of early multimodal fusion, we create a convolutional LSTM network architecture that simultaneously processes both audio and visual inputs, and allows us to select the layer at which audio and visual information combines. Our results demonstrate that immediate fusion of audio and visual inputs in the initial C-LSTM layer results in higher performing networks that are more robust to the addition of white noise in both audio and visual inputs.", "date": "2021-01-20", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20210119-161629149", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20210119-161629149", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2011.07191", "primary_object": { "basename": "2011.07191.pdf", "url": "https://authors.library.caltech.edu/records/q6qth-h9r71/files/2011.07191.pdf" }, "resource_type": "monograph", "pub_year": "2021", "author_list": "Barnum, George; Talukder, Sabera; et el." }, { "id": "https://authors.library.caltech.edu/records/4ax9n-s2b88", "eprint_id": 106601, "eprint_status": "archive", "datestamp": "2023-08-19 21:59:08", "lastmod": "2023-10-20 23:37:14", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Kumar-Akash", "name": { "family": "Kumar", "given": "Akash" } }, { "id": "Singla-Adish", "name": { "family": "Singla", "given": "Adish" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Chen-Yuxin", "name": { "family": "Chen", "given": "Yuxin" } } ] }, "title": "Average-case Complexity of Teaching Convex Polytopes via Halfspace Queries", "ispublished": "unpub", "full_text_status": "public", "keywords": "Teaching dimension, homogeneous halfspaces, average-case complexity", "note": "\u00a9 2020 A. Kumar, A. Singla, Y. Yue & Y. Chen. \n\nWe thank Ali Sayyadi for the helpful discussions. This work was supported in part by fundings from PIMCO and Bloomberg.\n\nSubmitted - 2006.14677.pdf
", "abstract": "We examine the task of locating a target region among those induced by intersections of n halfspaces in R^d. This generic task connects to fundamental machine learning problems, such as training a perceptron and learning a \u03d5-separable dichotomy. We investigate the average teaching complexity of the task, i.e., the minimal number of samples (halfspace queries) required by a teacher to help a version-space learner in locating a randomly selected target. As our main result, we show that the average-case teaching complexity is \u0398(d), which is in sharp contrast to the worst-case teaching complexity of \u0398(n). If instead, we consider the average-case learning complexity, the bounds have a dependency on n as \u0398(n) for i.i.d. queries and \u0398(dlog(n)) for actively chosen queries by the learner. Our proof techniques are based on novel insights from computational geometry, which allow us to count the number of convex polytopes and faces in a Euclidean space depending on the arrangement of halfspaces. Our insights allow us to establish a tight bound on the average-case complexity for \u03d5-separable dichotomies, which generalizes the known O(d) bound on the average number of \"extreme patterns\" in the classical computational geometry literature (Cover, 1965).", "date": "2020-11-11", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20201111-071759033", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20201111-071759033", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "PIMCO" }, { "agency": "Bloomberg Data Science" } ] }, "doi": "10.48550/arXiv.2006.14677", "primary_object": { "basename": "2006.14677.pdf", "url": "https://authors.library.caltech.edu/records/4ax9n-s2b88/files/2006.14677.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Kumar, Akash; Singla, Adish; et el." }, { "id": "https://authors.library.caltech.edu/records/fv4kv-5ky22", "eprint_id": 106598, "eprint_status": "archive", "datestamp": "2023-08-19 22:21:02", "lastmod": "2023-10-20 23:37:04", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Ryou-Serim", "name": { "family": "Ryou", "given": "Serim" } }, { "id": "Maser-Michael-R", "name": { "family": "Maser", "given": "Michael R." }, "orcid": "0000-0001-7895-7804" }, { "id": "Cui-Alexander-Y", "name": { "family": "Cui", "given": "Alexander Y." } }, { "id": "DeLano-Travis-J", "name": { "family": "DeLano", "given": "Travis J." }, "orcid": "0000-0002-2052-611X" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Reisman-S-E", "name": { "family": "Reisman", "given": "Sarah E." }, "orcid": "0000-0001-8244-9300" } ] }, "title": "Graph Neural Networks for the Prediction of Substrate-Specific Organic Reaction Conditions", "ispublished": "unpub", "full_text_status": "public", "note": "\u00a9 2020 by the author(s). \n\nTo appear in the ICML 2020 Workshop on Graph Representation\nLearning and Beyond (GRLB). \n\nWe thank the reviewers for their insightful comments and Prof Pietro Perona for mentorship guidance and helpful discussions on this work. Fellowship support was provided by the NSF (M.R.M., T.J.D. Grant No. DGE-1144469). S.E.R. is a Heritage Medical Research Investigator. Financial support from the Research Corporation Cottrell Scholars Program is acknowledged.\n\nSubmitted - 2007.04275.pdf
", "abstract": "We present a systematic investigation using graph neural networks (GNNs) to model organic chemical reactions. To do so, we prepared a dataset collection of four ubiquitous reactions from the organic chemistry literature. We evaluate seven different GNN architectures for classification tasks pertaining to the identification of experimental reagents and conditions. We find that models are able to identify specific graph features that affect reaction conditions and lead to accurate predictions. The results herein show great promise in advancing molecular machine learning.", "date": "2020-11-11", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20201110-154207213", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20201110-154207213", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NSF Graduate Research Fellowship", "grant_number": "DGE-1144469" }, { "agency": "Heritage Medical Research Institute" }, { "agency": "Cottrell Scholar of Research Corporation" } ] }, "local_group": { "items": [ { "id": "Heritage-Medical-Research-Institute" } ] }, "doi": "10.48550/arXiv.2007.04275", "primary_object": { "basename": "2007.04275.pdf", "url": "https://authors.library.caltech.edu/records/fv4kv-5ky22/files/2007.04275.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Ryou, Serim; Maser, Michael R.; et el." }, { "id": "https://authors.library.caltech.edu/records/nnyhw-95c64", "eprint_id": 106583, "eprint_status": "archive", "datestamp": "2023-08-19 23:57:34", "lastmod": "2023-10-20 23:36:27", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Yu-Chenkai", "name": { "family": "Yu", "given": "Chenkai" } }, { "id": "Shi-Guanya", "name": { "family": "Shi", "given": "Guanya" } }, { "id": "Chung-Soon-Jo", "name": { "family": "Chung", "given": "Soon-Jo" }, "orcid": "0000-0002-6657-3907" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Wierman-A", "name": { "family": "Wierman", "given": "Adam" } } ] }, "title": "Competitive Control with Delayed Imperfect Information", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 2010.11637.pdf
", "abstract": "This paper studies the impact of imperfect information in online control with adversarial disturbances. In particular, we consider both delayed state feedback and inexact predictions of future disturbances. We introduce a greedy, myopic policy that yields a constant competitive ratio against the offline optimal policy with delayed feedback and inexact predictions. A special case of our result is a constant competitive policy for the case of exact predictions and no delay, a previously open problem. We also analyze the fundamental limits of online control with limited information by showing that our competitive ratio bounds for the greedy, myopic policy in the adversarial setting match (up to lower-order terms) lower bounds in the stochastic setting.", "date": "2020-11-10", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20201110-082106076", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20201110-082106076", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "local_group": { "items": [ { "id": "GALCIT" } ] }, "doi": "10.48550/arXiv.2010.11637", "primary_object": { "basename": "2010.11637.pdf", "url": "https://authors.library.caltech.edu/records/nnyhw-95c64/files/2010.11637.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Yu, Chenkai; Shi, Guanya; et el." }, { "id": "https://authors.library.caltech.edu/records/f8x2p-rtc42", "eprint_id": 106584, "eprint_status": "archive", "datestamp": "2023-08-19 23:56:46", "lastmod": "2023-10-20 22:52:25", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Marino-Joseph-L", "name": { "family": "Marino", "given": "Joseph" }, "orcid": "0000-0001-6387-8062" }, { "id": "Pich\u00e9-Alexandre", "name": { "family": "Pich\u00e9", "given": "Alexandre" } }, { "id": "Ialongo-Alessandro-Davide", "name": { "family": "Ialongo", "given": "Alessandro Davide" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Iterative Amortized Policy Optimization", "ispublished": "unpub", "full_text_status": "public", "note": "JM acknowledges Scott Fujimoto for helpful discussions. This work was funded in part by NSF #1918839 and Beyond Limits. JM is currently employed by Google DeepMind. The authors declare no other competing interests related to this work.\n\nAccepted Version - 2010.10670.pdf
", "abstract": "Policy networks are a central feature of deep reinforcement learning (RL) algorithms for continuous control, enabling the estimation and sampling of high-value actions. From the variational inference perspective on RL, policy networks, when employed with entropy or KL regularization, are a form of amortized optimization, optimizing network parameters rather than the policy distributions directly. However, this direct amortized mapping can empirically yield suboptimal policy estimates. Given this perspective, we consider the more flexible class of iterative amortized optimizers. We demonstrate that the resulting technique, iterative amortized policy optimization, yields performance improvements over conventional direct amortization methods on benchmark continuous control tasks.", "date": "2020-11-10", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20201110-082336091", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20201110-082336091", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NSF", "grant_number": "CCF-1918839" }, { "agency": "Beyond Limits" } ] }, "doi": "10.48550/arXiv.2010.10670", "primary_object": { "basename": "2010.10670.pdf", "url": "https://authors.library.caltech.edu/records/f8x2p-rtc42/files/2010.10670.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Marino, Joseph; Pich\u00e9, Alexandre; et el." }, { "id": "https://authors.library.caltech.edu/records/hg146-9p613", "eprint_id": 106585, "eprint_status": "archive", "datestamp": "2023-08-19 22:31:43", "lastmod": "2023-10-20 23:36:30", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Shah-Ameesh", "name": { "family": "Shah", "given": "Ameesh" } }, { "id": "Zhan-Eric", "name": { "family": "Zhan", "given": "Eric" } }, { "id": "Sun-Jennifer-J", "name": { "family": "Sun", "given": "Jennifer J." }, "orcid": "0000-0002-0906-6589" }, { "id": "Verma-Abhinav", "name": { "family": "Verma", "given": "Abhinav" }, "orcid": "0000-0002-9820-8285" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Chaudhuri-Swarat", "name": { "family": "Chaudhuri", "given": "Swarat" } } ] }, "title": "Learning Differentiable Programs with Admissible Neural Heuristics", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 2007.12101.pdf
", "abstract": "We study the problem of learning differentiable functions expressed as programs in a domain-specific language. Such programmatic models can offer benefits such as composability and interpretability; however, learning them requires optimizing over a combinatorial space of program \"architectures\". We frame this optimization problem as a search in a weighted graph whose paths encode top-down derivations of program syntax. Our key innovation is to view various classes of neural networks as continuous relaxations over the space of programs, which can then be used to complete any partial program. This relaxed program is differentiable and can be trained end-to-end, and the resulting training loss is an approximately admissible heuristic that can guide the combinatorial search. We instantiate our approach on top of the A-star algorithm and an iteratively deepened branch-and-bound search, and use these algorithms to learn programmatic classifiers in three sequence classification tasks. Our experiments show that the algorithms outperform state-of-the-art methods for program learning, and that they discover programmatic classifiers that yield natural interpretations and achieve competitive accuracy.", "date": "2020-11-10", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20201110-085241409", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20201110-085241409", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2007.12101", "primary_object": { "basename": "2007.12101.pdf", "url": "https://authors.library.caltech.edu/records/hg146-9p613/files/2007.12101.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Shah, Ameesh; Zhan, Eric; et el." }, { "id": "https://authors.library.caltech.edu/records/hzymj-wxd44", "eprint_id": 106490, "eprint_status": "archive", "datestamp": "2023-08-19 21:55:14", "lastmod": "2023-10-20 23:32:41", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Prajapat-Manish", "name": { "family": "Prajapat", "given": "Manish" }, "orcid": "0000-0002-3867-4575" }, { "id": "Azizzadenesheli-Kamyar", "name": { "family": "Azizzadenesheli", "given": "Kamyar" }, "orcid": "0000-0001-8507-1868" }, { "id": "Liniger-Alexander", "name": { "family": "Liniger", "given": "Alexander" }, "orcid": "0000-0002-7858-7900" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Anandkumar-A", "name": { "family": "Anandkumar", "given": "Anima" }, "orcid": "0000-0002-6974-6797" } ] }, "title": "Competitive Policy Optimization", "ispublished": "unpub", "full_text_status": "public", "note": "The main body of this work took place when M. Prajapat was a visiting scholar at Caltech. The authors would like to thank Florian Sch\u00e4fer for his support. M. Prajapat is thankful to Zeno Karl Schindler foundation for providing him with a Master thesis grant. K. Azizzadenesheli is supported in part by Raytheon and Amazon Web Service. A. Anandkumar is supported in part by Bren endowed chair, DARPA PAIHR00111890035 and LwLL grants, Raytheon, Microsoft, Google, and Adobe faculty fellowships.\n\nSubmitted - 2006.10611.pdf
", "abstract": "A core challenge in policy optimization in competitive Markov decision processes is the design of efficient optimization methods with desirable convergence and stability properties. To tackle this, we propose competitive policy optimization (CoPO), a novel policy gradient approach that exploits the game-theoretic nature of competitive games to derive policy updates. Motivated by the competitive gradient optimization method, we derive a bilinear approximation of the game objective. In contrast, off-the-shelf policy gradient methods utilize only linear approximations, and hence do not capture interactions among the players. We instantiate CoPO in two ways:(i) competitive policy gradient, and (ii) trust-region competitive policy optimization. We theoretically study these methods, and empirically investigate their behavior on a set of comprehensive, yet challenging, competitive games. We observe that they provide stable optimization, convergence to sophisticated strategies, and higher scores when played against baseline policy gradient methods.", "date": "2020-11-06", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20201106-120215567", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20201106-120215567", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "Zeno Karl Schindler Foundation" }, { "agency": "Raytheon Company" }, { "agency": "Amazon Web Services" }, { "agency": "Bren Professor of Computing and Mathematical Sciences" }, { "agency": "Defense Advanced Research Projects Agency (DARPA)", "grant_number": "HR00111890035" }, { "agency": "Learning with Less Labels (LwLL)" }, { "agency": "Microsoft Faculty Fellowship" }, { "agency": "Google Faculty Research Award" }, { "agency": "Adobe" } ] }, "doi": "10.48550/arXiv.2006.10611", "primary_object": { "basename": "2006.10611.pdf", "url": "https://authors.library.caltech.edu/records/hzymj-wxd44/files/2006.10611.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Prajapat, Manish; Azizzadenesheli, Kamyar; et el." }, { "id": "https://authors.library.caltech.edu/records/tjpba-0za37", "eprint_id": 106482, "eprint_status": "archive", "datestamp": "2023-08-19 23:49:39", "lastmod": "2023-10-20 23:32:02", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Wang-Haoxuan-Shanghai-Jiao-Tong", "name": { "family": "Wang", "given": "Haoxuan" } }, { "id": "Liu-Anqi", "name": { "family": "Liu", "given": "Anqi" } }, { "id": "Yu-Zhiding", "name": { "family": "Yu", "given": "Zhiding" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Anandkumar-A", "name": { "family": "Anandkumar", "given": "Anima" } } ] }, "title": "Distributionally Robust Learning for Unsupervised Domain Adaptation", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 2010.05784.pdf
", "abstract": "We propose a distributionally robust learning (DRL) method for unsupervised domain adaptation (UDA) that scales to modern computer vision benchmarks. DRL can be naturally formulated as a competitive two-player game between a predictor and an adversary that is allowed to corrupt the labels, subject to certain constraints, and reduces to incorporating a density ratio between the source and target domains (under the standard log loss). This formulation motivates the use of two neural networks that are jointly trained - a discriminative network between the source and target domains for density-ratio estimation, in addition to the standard classification network. The use of a density ratio in DRL prevents the model from being overconfident on target inputs far away from the source domain. Thus, DRL provides conservative confidence estimation in the target domain, even when the target labels are not available. This conservatism motivates the use of DRL in self-training for sample selection, and we term the approach distributionally robust self-training (DRST). In our experiments, DRST generates more calibrated probabilities and achieves state-of-the-art self-training accuracy on benchmark datasets. We demonstrate that DRST captures shape features more effectively, and reduces the extent of distributional shift during self-training.", "date": "2020-11-06", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20201106-120148344", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20201106-120148344", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2010.05784", "primary_object": { "basename": "2010.05784.pdf", "url": "https://authors.library.caltech.edu/records/tjpba-0za37/files/2010.05784.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Wang, Haoxuan; Liu, Anqi; et el." }, { "id": "https://authors.library.caltech.edu/records/vk7vz-zp479", "eprint_id": 104236, "eprint_status": "archive", "datestamp": "2023-08-19 21:51:40", "lastmod": "2023-10-20 19:12:00", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Yu-Chenkai", "name": { "family": "Yu", "given": "Chenkai" }, "orcid": "0000-0001-8683-7773" }, { "id": "Shi-Guanya", "name": { "family": "Shi", "given": "Guanya" }, "orcid": "0000-0002-9075-3705" }, { "id": "Chung-Soon-Jo", "name": { "family": "Chung", "given": "Soon-Jo" }, "orcid": "0000-0002-6657-3907" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Wierman-A", "name": { "family": "Wierman", "given": "Adam" }, "orcid": "0000-0002-5923-0199" } ] }, "title": "The Power of Predictions in Online Control", "ispublished": "unpub", "full_text_status": "public", "note": "This project was supported in part by funding from Raytheon, DARPA PAI, AitF-1637598 and CNS-1518941, with additional support for Guanya Shi provided by the Simoudis Discovery Prize. \n\nWe see no ethical concerns related to the results in this paper.\n\nAccepted Version - 2006.07569.pdf
", "abstract": "We study the impact of predictions in online Linear Quadratic Regulator control with both stochastic and adversarial disturbances in the dynamics. In both settings, we characterize the optimal policy and derive tight bounds on the minimum cost and dynamic regret. Perhaps surprisingly, our analysis shows that the conventional greedy MPC approach is a near-optimal policy in both stochastic and adversarial settings. Specifically, for length-T problems, MPC requires only O(logT) predictions to reach O(1) dynamic regret, which matches (up to lower-order terms) our lower bound on the required prediction horizon for constant regret.", "date": "2020-07-07", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20200707-094715120", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20200707-094715120", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "Raytheon Company" }, { "agency": "Defense Advanced Research Projects Agency (DARPA)" }, { "agency": "NSF", "grant_number": "CCF-1637598" }, { "agency": "NSF", "grant_number": "CNS-1518941" }, { "agency": "Simoudis Discovery Prize" } ] }, "local_group": { "items": [ { "id": "GALCIT" } ] }, "doi": "10.48550/arXiv.2006.07569", "primary_object": { "basename": "2006.07569.pdf", "url": "https://authors.library.caltech.edu/records/vk7vz-zp479/files/2006.07569.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Yu, Chenkai; Shi, Guanya; et el." }, { "id": "https://authors.library.caltech.edu/records/7sv9q-5fv63", "eprint_id": 103473, "eprint_status": "archive", "datestamp": "2023-08-19 20:39:18", "lastmod": "2023-10-20 16:24:00", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Song-Jialin", "name": { "family": "Song", "given": "Jialin" } }, { "id": "Lanka-Ravi", "name": { "family": "Lanka", "given": "Ravi" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Dilkina-Bistra", "name": { "family": "Dilkina", "given": "Bistra" }, "orcid": "0000-0002-6784-473X" } ] }, "title": "A General Large Neighborhood Search Framework for Solving Integer Programs", "ispublished": "unpub", "full_text_status": "public", "note": "We thank the anonymous reviewers for their suggestions for improvements. Dilkina was supported partially by NSF #1763108, DARPA, DHS Center of Excellence \"Critical Infrastructure Resilience Institute\", and Microsoft. This research was also supported in part by funding from NSF #1645832, Raytheon, Beyond Limits, and JPL.\n\nAccepted Version - 2004.00422v3.pdf
", "abstract": "This paper studies how to design abstractions of large-scale combinatorial optimization problems that can leverage existing state-of-the-art solvers in general purpose ways, and that are amenable to data-driven design. The goal is to arrive at new approaches that can reliably outperform existing solvers in wall-clock time. We focus on solving integer programs, and ground our approach in the large neighborhood search (LNS) paradigm, which iteratively chooses a subset of variables to optimize while leaving the remainder fixed. The appeal of LNS is that it can easily use any existing solver as a subroutine, and thus can inherit the benefits of carefully engineered heuristic approaches and their software implementations. We also show that one can learn a good neighborhood selector from training data. Through an extensive empirical validation, we demonstrate that our LNS framework can significantly outperform, in wall-clock time, compared to state-of-the-art commercial solvers such as Gurobi.", "date": "2020-05-26", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20200526-151215262", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20200526-151215262", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NSF", "grant_number": "CMMI-1763108" }, { "agency": "Defense Advanced Research Projects Agency (DARPA)" }, { "agency": "Department of Homeland Security" }, { "agency": "Microsoft Research" }, { "agency": "NSF", "grant_number": "CNS-1645832" }, { "agency": "Raytheon Company" }, { "agency": "Beyond Limits" }, { "agency": "JPL" } ] }, "doi": "10.48550/arXiv.2004.00422", "primary_object": { "basename": "2004.00422v3.pdf", "url": "https://authors.library.caltech.edu/records/7sv9q-5fv63/files/2004.00422v3.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Song, Jialin; Lanka, Ravi; et el." }, { "id": "https://authors.library.caltech.edu/records/aptqa-b9c21", "eprint_id": 101304, "eprint_status": "archive", "datestamp": "2023-08-19 19:58:13", "lastmod": "2023-10-19 22:36:39", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Park-Jung-Yeon", "name": { "family": "Park", "given": "Jung Yeon" } }, { "id": "Carr-K-T", "name": { "family": "Carr", "given": "Kenneth Theo" } }, { "id": "Zhang-Stephan", "name": { "family": "Zhang", "given": "Stephan" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Yu-Rose", "name": { "family": "Yu", "given": "Rose" } } ] }, "title": "Multiresolution Tensor Learning for Efficient and Interpretable Spatial Analysis", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 2002.05578.pdf
", "abstract": "Efficient and interpretable spatial analysis is crucial in many fields such as geology, sports, and climate science. Large-scale spatial data often contains complex higher-order correlations across features and locations. While tensor latent factor models can describe higher-order correlations, they are inherently computationally expensive to train. Furthermore, for spatial analysis, these models should not only be predictive but also be spatially coherent. However, latent factor models are sensitive to initialization and can yield inexplicable results. We develop a novel Multi-resolution Tensor Learning (MRTL) algorithm for efficiently learning interpretable spatial patterns. MRTL initializes the latent factors from an approximate full-rank tensor model for improved interpretability and progressively learns from a coarse resolution to the fine resolution for an enormous computation speedup. We also prove the theoretical convergence and computational complexity of MRTL. When applied to two real-world datasets, MRTL demonstrates 4 ~ 5 times speedup compared to a fixed resolution while yielding accurate and interpretable models.", "date": "2020-02-14", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20200214-105610460", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20200214-105610460", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.2002.05578", "primary_object": { "basename": "2002.05578.pdf", "url": "https://authors.library.caltech.edu/records/aptqa-b9c21/files/2002.05578.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Park, Jung Yeon; Carr, Kenneth Theo; et el." }, { "id": "https://authors.library.caltech.edu/records/wyfv9-end27", "eprint_id": 101302, "eprint_status": "archive", "datestamp": "2023-08-19 19:56:10", "lastmod": "2023-10-20 22:18:22", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Bernstein-Jeremy-D", "name": { "family": "Bernstein", "given": "Jeremy" }, "orcid": "0000-0001-9110-7476" }, { "id": "Vahdat-Arash", "name": { "family": "Vahdat", "given": "Arash" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Liu-Ming-Yu", "name": { "family": "Liu", "given": "Ming-Yu" }, "orcid": "0000-0002-2951-2398" } ] }, "title": "On the distance between two neural networks and the stability of learning", "ispublished": "unpub", "full_text_status": "public", "note": "The authors would like to thank Dillon Huff, Jeffrey Pennington and Florian Schaefer for useful conversations. They made heavy use of a codebase built by Jiahui Yu. They are much obliged to Sivakumar Arayandi Thottakara, Jan Kautz, Sabu Nadarajan and Nithya Natesan for infrastructure support. JB is supported by an NVIDIA fellowship.\n\nAccepted Version - 2002.03432.pdf
", "abstract": "This paper relates parameter distance to gradient breakdown for a broad class of nonlinear compositional functions. The analysis leads to a new distance function called deep relative trust and a descent lemma for neural networks. Since the resulting learning rule seems to require little to no learning rate tuning, it may unlock a simpler workflow for training deeper and more complex neural networks. The Python code used in this paper is here: https://github.com/jxbz/fromage", "date": "2020-02-14", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20200214-105602886", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20200214-105602886", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NVIDIA" } ] }, "doi": "10.48550/arXiv.2002.03432", "primary_object": { "basename": "2002.03432.pdf", "url": "https://authors.library.caltech.edu/records/wyfv9-end27/files/2002.03432.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Bernstein, Jeremy; Vahdat, Arash; et el." }, { "id": "https://authors.library.caltech.edu/records/xrywz-v0w33", "eprint_id": 101303, "eprint_status": "archive", "datestamp": "2023-08-19 19:58:08", "lastmod": "2023-10-19 22:36:37", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Shi-Guanya", "name": { "family": "Shi", "given": "Guanya" }, "orcid": "0000-0002-9075-3705" }, { "id": "Lin-Yiheng", "name": { "family": "Lin", "given": "Yiheng" }, "orcid": "0000-0001-6524-2877" }, { "id": "Chung-Soon-Jo", "name": { "family": "Chung", "given": "Soon-Jo" }, "orcid": "0000-0002-6657-3907" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Wierman-A", "name": { "family": "Wierman", "given": "Adam" }, "orcid": "0000-0002-5923-0199" } ] }, "title": "Online Optimization with Memory and Competitive Control", "ispublished": "unpub", "full_text_status": "public", "note": "This project was supported in part by funding from Raytheon, DARPA PAI, AitF-1637598 and CNS-1518941, with additional support for Guanya Shi provided by the Simoudis Discovery Prize. \n\nWe see no ethical concerns related to the results in this paper.\n\nAccepted Version - 2002.05318.pdf
", "abstract": "This paper presents competitive algorithms for a novel class of online optimization problems with memory. We consider a setting where the learner seeks to minimize the sum of a hitting cost and a switching cost that depends on the previous p decisions. This setting generalizes Smoothed Online Convex Optimization. The proposed approach, Optimistic Regularized Online Balanced Descent, achieves a constant, dimension-free competitive ratio. Further, we show a connection between online optimization with memory and online control with adversarial disturbances. This connection, in turn, leads to a new constant-competitive policy for a rich class of online control problems.", "date": "2020-02-14", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20200214-105606928", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20200214-105606928", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "Raytheon Company" }, { "agency": "Defense Advanced Research Projects Agency (DARPA)" }, { "agency": "NSF", "grant_number": "CCF-1637598" }, { "agency": "NSF", "grant_number": "CNS-1518941" }, { "agency": "Simoudis Discovery Prize" } ] }, "local_group": { "items": [ { "id": "GALCIT" } ] }, "doi": "10.48550/arXiv.2002.05318", "primary_object": { "basename": "2002.05318.pdf", "url": "https://authors.library.caltech.edu/records/xrywz-v0w33/files/2002.05318.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Shi, Guanya; Lin, Yiheng; et el." }, { "id": "https://authors.library.caltech.edu/records/6vcht-tev43", "eprint_id": 100590, "eprint_status": "archive", "datestamp": "2023-08-19 18:45:41", "lastmod": "2023-10-18 21:38:02", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Voloshin-C", "name": { "family": "Voloshin", "given": "Cameron" } }, { "id": "Le-Hoang-M", "name": { "family": "Le", "given": "Hoang M." } }, { "id": "Jiang-Nan", "name": { "family": "Jiang", "given": "Nan" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Empirical Study of Off-Policy Policy Evaluation for Reinforcement Learning", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 1911.06854.pdf
", "abstract": "Off-policy policy evaluation (OPE) is the problem of estimating the online performance of a policy using only pre-collected historical data generated by another policy. Given the increasing interest in deploying learning-based methods for safety-critical applications, many recent OPE methods have recently been proposed. Due to disparate experimental conditions from recent literature, the relative performance of current OPE methods is not well understood. In this work, we present the first comprehensive empirical analysis of a broad suite of OPE methods. Based on thousands of experiments and detailed empirical analyses, we offer a summarized set of guidelines for effectively using OPE in practice, and suggest directions for future research.", "date": "2020-01-09", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20200109-100747650", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20200109-100747650", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.1911.06854", "primary_object": { "basename": "1911.06854.pdf", "url": "https://authors.library.caltech.edu/records/6vcht-tev43/files/1911.06854.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Voloshin, Cameron; Le, Hoang M.; et el." }, { "id": "https://authors.library.caltech.edu/records/ndz25-g8638", "eprint_id": 100592, "eprint_status": "archive", "datestamp": "2023-08-19 18:12:39", "lastmod": "2023-10-18 21:38:08", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Zhan-Eric", "name": { "family": "Zhan", "given": "Eric" } }, { "id": "Tseng-Albert", "name": { "family": "Tseng", "given": "Albert" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Swaminathan-A", "name": { "family": "Swaminathan", "given": "Adith" }, "orcid": "0000-0001-9935-6530" }, { "id": "Hausknecht-M", "name": { "family": "Hausknecht", "given": "Matthew" } } ] }, "title": "Learning Calibratable Policies using Programmatic Style-Consistency", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 1910.01179.pdf
", "abstract": "We study the important and challenging problem of controllable generation of long-term sequential behaviors. Solutions to this problem would impact many applications, such as calibrating behaviors of AI agents in games or predicting player trajectories in sports. In contrast to the well-studied areas of controllable generation of images, text, and speech, there are significant challenges that are unique to or exacerbated by generating long-term behaviors: how should we specify the factors of variation to control, and how can we ensure that the generated temporal behavior faithfully demonstrates diverse styles? In this paper, we leverage large amounts of raw behavioral data to learn policies that can be calibrated to generate a diverse range of behavior styles (e.g., aggressive versus passive play in sports). Inspired by recent work on leveraging programmatic labeling functions, we present a novel framework that combines imitation learning with data programming to learn style-calibratable policies. Our primary technical contribution is a formal notion of style-consistency as a learning objective, and its integration with conventional imitation learning approaches. We evaluate our framework using demonstrations from professional basketball players and agents in the MuJoCo physics environment, and show that our learned policies can be accurately calibrated to generate interesting behavior styles in both domains.", "date": "2020-01-09", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20200109-101924329", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20200109-101924329", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.1910.01179", "primary_object": { "basename": "1910.01179.pdf", "url": "https://authors.library.caltech.edu/records/ndz25-g8638/files/1910.01179.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Zhan, Eric; Tseng, Albert; et el." }, { "id": "https://authors.library.caltech.edu/records/h450g-t2m88", "eprint_id": 100578, "eprint_status": "archive", "datestamp": "2023-08-19 18:43:26", "lastmod": "2023-10-18 21:37:09", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Liu-Anqi", "name": { "family": "Liu", "given": "Anqi" } }, { "id": "Liu-Hao", "name": { "family": "Liu", "given": "Hao" } }, { "id": "Anandkumar-A", "name": { "family": "Anandkumar", "given": "Anima" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Triply Robust Off-Policy Evaluation", "ispublished": "unpub", "full_text_status": "public", "note": "Prof. Anandkumar is supported by Bren endowed Chair, faculty awards from Microsoft, Google, and Adobe, DARPA PAI and LwLL grants. Anqi Liu is a PIMCO postdoctoral fellow at Caltech.\n\nSubmitted - 1911.05811.pdf
", "abstract": "We propose a robust regression approach to off-policy evaluation (OPE) for contextual bandits. We frame OPE as a covariate-shift problem and leverage modern robust regression tools. Ours is a general approach that can be used to augment any existing OPE method that utilizes the direct method. When augmenting doubly robust methods, we call the resulting method Triply Robust. We prove upper bounds on the resulting bias and variance, as well as derive novel minimax bounds based on robust minimax analysis for covariate shift. Our robust regression method is compatible with deep learning, and is thus applicable to complex OPE settings that require powerful function approximators. Finally, we demonstrate superior empirical performance across the standard OPE benchmarks, especially in the case where the logging policy is unknown and must be estimated from data.", "date": "2020-01-09", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20200109-085907638", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20200109-085907638", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "Bren Professor of Computing and Mathematical Sciences" }, { "agency": "Microsoft" }, { "agency": "Google" }, { "agency": "Adobe" }, { "agency": "Defense Advanced Research Projects Agency (DARPA)" }, { "agency": "Caltech PIMCO Graduate Fellowship" } ] }, "doi": "10.48550/arXiv.1911.05811", "primary_object": { "basename": "1911.05811.pdf", "url": "https://authors.library.caltech.edu/records/h450g-t2m88/files/1911.05811.pdf" }, "resource_type": "monograph", "pub_year": "2020", "author_list": "Liu, Anqi; Liu, Hao; et el." }, { "id": "https://authors.library.caltech.edu/records/kws8t-j2v54", "eprint_id": 98459, "eprint_status": "archive", "datestamp": "2023-08-19 16:42:21", "lastmod": "2023-10-18 17:23:02", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Song-Jialin", "name": { "family": "Song", "given": "Jialin" } }, { "id": "Lanka-R", "name": { "family": "Lanka", "given": "Ravi" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Ono-Masahiro", "name": { "family": "Ono", "given": "Masahiro" } } ] }, "title": "Co-training for Policy Learning", "ispublished": "unpub", "full_text_status": "public", "note": "The work was funded in part by NSF awards #1637598 & #1645832, and support from Raytheon and Northrop Grumman. This research was also conducted in part at the Jet Propulsion Lab, California Insitute of Technology under a contract with the National Aeronautics and Space Administration.\n\nSubmitted - 1907.04484.pdf
", "abstract": "We study the problem of learning sequential decision-making policies in settings with multiple state-action representations. Such settings naturally arise in many domains, such as planning (e.g., multiple integer programming formulations) and various combinatorial optimization problems (e.g., those with both integer programming and graph-based formulations). Inspired by the classical co-training framework for classification, we study the problem of co-training for policy learning. We present sufficient conditions under which learning from two views can improve upon learning from a single view alone. Motivated by these theoretical insights, we present a meta-algorithm for co-training for sequential decision making. Our framework is compatible with both reinforcement learning and imitation learning. We validate the effectiveness of our approach across a wide range of tasks, including discrete/continuous control and combinatorial optimization.", "date": "2019-09-05", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20190905-154310582", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20190905-154310582", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NSF", "grant_number": "CCF-1637598" }, { "agency": "NSF", "grant_number": "CNS-1645832" }, { "agency": "Raytheon Company" }, { "agency": "Northrop Grumman Corporation" }, { "agency": "NASA/JPL/Caltech" } ] }, "doi": "10.48550/arXiv.1907.04484", "primary_object": { "basename": "1907.04484.pdf", "url": "https://authors.library.caltech.edu/records/kws8t-j2v54/files/1907.04484.pdf" }, "resource_type": "monograph", "pub_year": "2019", "author_list": "Song, Jialin; Lanka, Ravi; et el." }, { "id": "https://authors.library.caltech.edu/records/607xd-w9259", "eprint_id": 98458, "eprint_status": "archive", "datestamp": "2023-08-19 16:18:06", "lastmod": "2023-10-18 17:23:00", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Liu-Anqi", "name": { "family": "Liu", "given": "Anqi" } }, { "id": "Shi-Guanya", "name": { "family": "Shi", "given": "Guanya" } }, { "id": "Chung-Soon-Jo", "name": { "family": "Chung", "given": "Soon-Jo" }, "orcid": "0000-0002-6657-3907" }, { "id": "Anandkumar-A", "name": { "family": "Anandkumar", "given": "Anima" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Robust Regression for Safe Exploration in Control", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 1906.05819.pdf
", "abstract": "We study the problem of safe learning and exploration in sequential control problems. The goal is to safely collect data samples from an operating environment to learn an optimal controller. A central challenge in this setting is how to quantify uncertainty in order to choose provably-safe actions that allow us to collect useful data and reduce uncertainty, thereby achieving both improved safety and optimality. To address this challenge, we present a deep robust regression model that is trained to directly predict the uncertainty bounds for safe exploration. We then show how to integrate our robust regression approach with model-based control methods by learning a dynamic model with robustness bounds. We derive generalization bounds under domain shifts for learning and connect them with safety and stability bounds in control. We demonstrate empirically that our robust regression approach can outperform conventional Gaussian process (GP) based safe exploration in settings where it is difficult to specify a good GP prior.", "date": "2019-09-05", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20190905-154307157", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20190905-154307157", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.1906.05819", "primary_object": { "basename": "1906.05819.pdf", "url": "https://authors.library.caltech.edu/records/607xd-w9259/files/1906.05819.pdf" }, "resource_type": "monograph", "pub_year": "2019", "author_list": "Liu, Anqi; Shi, Guanya; et el." }, { "id": "https://authors.library.caltech.edu/records/egwzd-zz505", "eprint_id": 94640, "eprint_status": "archive", "datestamp": "2023-08-19 02:46:37", "lastmod": "2023-10-20 18:08:08", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Sui-Yanan", "name": { "family": "Sui", "given": "Yanan" }, "orcid": "0000-0002-9480-627X" }, { "id": "Zhuang-Vincent", "name": { "family": "Zhuang", "given": "Vincent" } }, { "id": "Burdick-J-W", "name": { "family": "Burdick", "given": "Joel W." } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Multi-dueling Bandits with Dependent Arms", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 1705.00253.pdf
", "abstract": "The dueling bandits problem is an online learning framework for learning from pairwise preference feedback, and is particularly well-suited for modeling settings that elicit subjective or implicit human feedback. In this paper, we study the problem of multi-dueling bandits with dependent arms, which extends the original dueling bandits setting by simultaneously dueling multiple arms as well as modeling dependencies between arms. These extensions capture key characteristics found in many real-world applications, and allow for the opportunity to develop significantly more efficient algorithms than were possible in the original setting. We propose the selfsparring algorithm, which reduces the multi-dueling bandits problem to a conventional bandit setting that can be solved using a stochastic bandit algorithm such as Thompson Sampling, and can naturally model dependencies using a Gaussian process prior. We present a no-regret analysis for multi-dueling setting, and demonstrate the effectiveness of our algorithm empirically on a wide range of simulation settings.", "date": "2019-04-10", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20190410-120658254", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20190410-120658254", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.1705.00253", "primary_object": { "basename": "1705.00253.pdf", "url": "https://authors.library.caltech.edu/records/egwzd-zz505/files/1705.00253.pdf" }, "resource_type": "monograph", "pub_year": "2019", "author_list": "Sui, Yanan; Zhuang, Vincent; et el." }, { "id": "https://authors.library.caltech.edu/records/da6hd-4mf72", "eprint_id": 92675, "eprint_status": "archive", "datestamp": "2023-08-19 04:04:29", "lastmod": "2023-10-20 16:18:46", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Sui-Yanan", "name": { "family": "Sui", "given": "Yanan" }, "orcid": "0000-0002-9480-627X" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Burdick-J-W", "name": { "family": "Burdick", "given": "Joel W." } } ] }, "title": "Correlational Dueling Bandits with Application to Clinical Treatment in Large Decision Spaces", "ispublished": "unpub", "full_text_status": "public", "note": "This research was supported in part by Caltech/JPL PDF IAMS100224, NIH-U01-EB007615-08, NIH-U01-EB015521-05, and a gift from Northrop Grumman.\n\nSubmitted - 1707.02375.pdf
", "abstract": "We consider sequential decision making under uncertainty, where the goal is to optimize over a large decision space using noisy comparative feedback. This problem can be formulated as a K-armed Dueling Bandits problem where K is the total number of decisions. When K is very large, existing dueling bandits algorithms suffer huge cumulative regret before converging on the optimal arm. This paper studies the dueling bandits problem with a large number of arms that exhibit a low-dimensional correlation structure. Our problem is motivated by a clinical decision making process in large decision space. We propose an efficient algorithm CorrDuel which optimizes the exploration/exploitation tradeoff in this large decision space of clinical treatments. More broadly, our approach can be applied to other sequential decision problems with large and structured decision spaces. We derive regret bounds, and evaluate performance in simulation experiments as well as on a live clinical trial of therapeutic spinal cord stimulation. To our knowledge, this marks the first time an online learning algorithm was applied towards spinal cord injury treatments. Our experimental results show the effectiveness and efficiency of our approach.", "date": "2019-02-05", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20190205-133559444", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20190205-133559444", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "JPL President and Director's Fund", "grant_number": "IAMS100224" }, { "agency": "NIH", "grant_number": "U01-EB007615-08" }, { "agency": "NIH", "grant_number": "U01-EB015521-05" }, { "agency": "Northrop Grumman" } ] }, "doi": "10.48550/arXiv.1707.02375", "primary_object": { "basename": "1707.02375.pdf", "url": "https://authors.library.caltech.edu/records/da6hd-4mf72/files/1707.02375.pdf" }, "resource_type": "monograph", "pub_year": "2019", "author_list": "Sui, Yanan; Yue, Yisong; et el." }, { "id": "https://authors.library.caltech.edu/records/687er-8c179", "eprint_id": 92670, "eprint_status": "archive", "datestamp": "2023-08-19 08:18:48", "lastmod": "2023-10-20 16:18:29", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Dathathri-S", "name": { "family": "Dathathri", "given": "Sumanth" } }, { "id": "Zheng-Stephan", "name": { "family": "Zheng", "given": "Stephan" } }, { "id": "Murray-R-M", "name": { "family": "Murray", "given": "Richard M." }, "orcid": "0000-0002-5785-7481" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Detecting Adversarial Examples via Neural Fingerprinting", "ispublished": "unpub", "full_text_status": "public", "note": "This work is supported in part by NSF grants #1564330, #1637598, #1545126; STARnet, a Semiconductor Research Corporation program, sponsored by MARCO and DARPA; and gifts from Bloomberg and Northrop Grumman. The authors would like to thank Xingjun Ma for providing the relevant baseline numbers for comparison.\n\nSubmitted - 1803.03870.pdf
", "abstract": "Deep neural networks are vulnerable to adversarial examples, which dramatically alter model output using small input changes. We propose Neural Fingerprinting, a simple, yet effective method to detect adversarial examples by verifying whether model behavior is consistent with a set of secret fingerprints, inspired by the use of biometric and cryptographic signatures. The benefits of our method are that 1) it is fast, 2) it is prohibitively expensive for an attacker to reverse-engineer which fingerprints were used, and 3) it does not assume knowledge of the adversary. In this work, we pose a formal framework to analyze fingerprints under various threat models, and characterize Neural Fingerprinting for linear models. For complex neural networks, we empirically demonstrate that Neural Fingerprinting significantly improves on state-of-the-art detection mechanisms by detecting the strongest known adversarial attacks with 98-100% AUC-ROC scores on the MNIST, CIFAR-10 and MiniImagenet (20 classes) datasets. In particular, the detection accuracy of Neural Fingerprinting generalizes well to unseen test-data under various black- and whitebox threat models, and is robust over a wide range of hyperparameters and choices of fingerprints.", "date": "2019-02-05", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20190205-112328842", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20190205-112328842", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NSF", "grant_number": "IIS-1564330" }, { "agency": "NSF", "grant_number": "CCF-1637598" }, { "agency": "NSF", "grant_number": "CNS-1545126" }, { "agency": "STARnet" }, { "agency": "Semiconductor Research Corporation" }, { "agency": "Microelectronics Advanced Research Corporation (MARCO)" }, { "agency": "Defense Advanced Research Projects Agency (DARPA)" }, { "agency": "Bloomberg Data Science" }, { "agency": "Northrop Grumman" } ] }, "doi": "10.48550/arXiv.1803.03870", "primary_object": { "basename": "1803.03870.pdf", "url": "https://authors.library.caltech.edu/records/687er-8c179/files/1803.03870.pdf" }, "resource_type": "monograph", "pub_year": "2019", "author_list": "Dathathri, Sumanth; Zheng, Stephan; et el." }, { "id": "https://authors.library.caltech.edu/records/53p9x-0y271", "eprint_id": 92673, "eprint_status": "archive", "datestamp": "2023-08-19 05:34:02", "lastmod": "2023-10-20 16:18:39", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Sha-Long", "name": { "family": "Sha", "given": "Long" } }, { "id": "Lucey-P", "name": { "family": "Lucey", "given": "Patrick" } }, { "id": "Zheng-Stephan", "name": { "family": "Zheng", "given": "Stephan" } }, { "id": "Kim-Taehwan", "name": { "family": "Kim", "given": "Taehwan" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Sridharan-S", "name": { "family": "Sridharan", "given": "Sridha" } } ] }, "title": "Fine-Grained Retrieval of Sports Plays using Tree-Based Alignment of Trajectories", "ispublished": "unpub", "full_text_status": "public", "keywords": "Retrieval and ranking, Multi-Agent Spatiotemporal Data, Data\nAlignment", "note": "\u00a9 2018 held by the owner/author(s).\n\nSubmitted - 1710.02255.pdf
", "abstract": "We propose a novel method for effective retrieval of multi-agent spatiotemporal tracking data. Retrieval of spatiotemporal tracking data offers several unique challenges compared to conventional text-based retrieval settings. Most notably, the data is fine-grained meaning that the specific location of agents is important in describing behavior. Additionally, the data often contains tracks of multiple agents (e.g., multiple players in a sports game), which generally leads to a permutational alignment problem when performing relevance estimation. Due to the frequent position swap of agents, it is difficult to maintain the correspondence of agents, and such issues make the pairwise comparison problematic for multi-agent spatiotemporal data. To address this issue, we propose a tree-based method to estimate the relevance between multi-agent spatiotemporal tracks. It uses a hierarchical structure to perform multi-agent data alignment and partitioning in a coarse-to-fine fashion. We validate our approach via user studies with domain experts. Our results show that our method boosts performance in retrieving similar sports plays -- especially in interactive situations where the user selects a subset of trajectories compared to current state-of-the-art methods.", "date": "2019-02-05", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20190205-113745110", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20190205-113745110", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.1710.02255", "primary_object": { "basename": "1710.02255.pdf", "url": "https://authors.library.caltech.edu/records/53p9x-0y271/files/1710.02255.pdf" }, "resource_type": "monograph", "pub_year": "2019", "author_list": "Sha, Long; Lucey, Patrick; et el." }, { "id": "https://authors.library.caltech.edu/records/jc24z-mgw41", "eprint_id": 92669, "eprint_status": "archive", "datestamp": "2023-08-19 08:23:39", "lastmod": "2023-10-20 16:18:26", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Zhan-Eric", "name": { "family": "Zhan", "given": "Eric" } }, { "id": "Zheng-Stephan", "name": { "family": "Zheng", "given": "Stephan" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Sha-Long", "name": { "family": "Sha", "given": "Long" } }, { "id": "Lucey-P", "name": { "family": "Lucey", "given": "Patrick" } } ] }, "title": "Generative Multi-Agent Behavioral Cloning", "ispublished": "unpub", "full_text_status": "public", "note": "This research is supported in part by NSF #1564330, NSF #1637598, and gifts from Bloomberg, Activision/Blizzard and Northrop Grumman. Dataset was provided by STATS: https://www.stats.com/data-science/.\n\nSubmitted - 1803.07612.pdf
", "abstract": "We propose and study the problem of generative multi-agent behavioral cloning, where the goal is to learn a generative, i.e., non-deterministic, multi-agent policy from pre-collected demonstration data. Building upon advances in deep generative models, we present a hierarchical policy framework that can tractably learn complex mappings from input states to distributions over multi-agent action spaces by introducing a hierarchy with macro-intent variables that encode long-term intent. In addition to synthetic settings, we show how to instantiate our framework to effectively model complex interactions between basketball players and generate realistic multi-agent trajectories of basketball gameplay over long time periods. We validate our approach using both quantitative and qualitative evaluations, including a user study comparison conducted with professional sports analysts.", "date": "2019-02-05", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20190205-111434225", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20190205-111434225", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NSF", "grant_number": "IIS-1564330" }, { "agency": "NSF", "grant_number": "CCF-1637598" }, { "agency": "Bloomberg Data Science" }, { "agency": "Activision/Blizzard" }, { "agency": "Northrop Grumman" } ] }, "doi": "10.48550/arXiv.1803.07612", "primary_object": { "basename": "1803.07612.pdf", "url": "https://authors.library.caltech.edu/records/jc24z-mgw41/files/1803.07612.pdf" }, "resource_type": "monograph", "pub_year": "2019", "author_list": "Zhan, Eric; Zheng, Stephan; et el." }, { "id": "https://authors.library.caltech.edu/records/g4q3m-mgf65", "eprint_id": 92668, "eprint_status": "archive", "datestamp": "2023-08-19 08:42:09", "lastmod": "2023-10-20 16:18:22", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Song-Jialin", "name": { "family": "Song", "given": "Jialin" } }, { "id": "Lanka-R", "name": { "family": "Lanka", "given": "Ravi" } }, { "id": "Zhao-Albert", "name": { "family": "Zhao", "given": "Albert" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Ono-Masahiro", "name": { "family": "Ono", "given": "Masahiro" } } ] }, "title": "Learning to Search via Retrospective Imitation", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 1804.00846.pdf
", "abstract": "We study the problem of learning a good search policy from demonstrations for combinatorial search spaces. We propose retrospective imitation learning, which, after initial training by an expert, improves itself by learning from its own retrospective solutions. That is, when the policy eventually reaches a feasible solution in a search tree after making mistakes and backtracks, it retrospectively constructs an improved search trace to the solution by removing backtracks, which is then used to further train the policy. A key feature of our approach is that it can iteratively scale up, or transfer, to larger problem sizes than the initial expert demonstrations, thus dramatically expanding its applicability beyond that of conventional imitation learning. We showcase the effectiveness of our approach on two tasks: synthetic maze solving, and integer program based risk-aware path planning.", "date": "2019-02-05", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20190205-111204454", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20190205-111204454", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.1804.00846", "primary_object": { "basename": "1804.00846.pdf", "url": "https://authors.library.caltech.edu/records/g4q3m-mgf65/files/1804.00846.pdf" }, "resource_type": "monograph", "pub_year": "2019", "author_list": "Song, Jialin; Lanka, Ravi; et el." }, { "id": "https://authors.library.caltech.edu/records/j9tmb-65s22", "eprint_id": 92672, "eprint_status": "archive", "datestamp": "2023-08-19 05:45:43", "lastmod": "2023-10-20 16:18:36", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Yu-Rose", "name": { "family": "Yu", "given": "Rose" } }, { "id": "Zheng-Stephan", "name": { "family": "Zheng", "given": "Stephan" } }, { "id": "Anandkumar-A", "name": { "family": "Anandkumar", "given": "Anima" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Long-term Forecasting using Tensor-Train RNNs", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 1711.00073.pdf
", "abstract": "We present Tensor-Train RNN (TT-RNN), a novel family of neural sequence architectures for multivariate forecasting in environments with nonlinear dynamics. Long-term forecasting in such systems is highly challenging, since there exist long-term temporal dependencies, higher-order correlations and sensitivity to error propagation. Our proposed tensor recurrent architecture addresses these issues by learning the nonlinear dynamics directly using higher order moments and high-order state transition functions. Furthermore, we decompose the higher-order structure using the tensor-train (TT) decomposition to reduce the number of parameters while preserving the model performance. We theoretically establish the approximation properties of Tensor-Train RNNs for general sequence inputs, and such guarantees are not available for usual RNNs. We also demonstrate significant long-term prediction improvements over general RNN and LSTM architectures on a range of simulated environments with nonlinear dynamics, as well on real-world climate and traffic data.", "date": "2019-02-05", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20190205-113450468", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20190205-113450468", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.1711.00073", "primary_object": { "basename": "1711.00073.pdf", "url": "https://authors.library.caltech.edu/records/j9tmb-65s22/files/1711.00073.pdf" }, "resource_type": "monograph", "pub_year": "2019", "author_list": "Yu, Rose; Zheng, Stephan; et el." }, { "id": "https://authors.library.caltech.edu/records/amrte-t0y10", "eprint_id": 92659, "eprint_status": "archive", "datestamp": "2023-08-19 12:45:37", "lastmod": "2023-10-20 15:57:19", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Song-Jialin", "name": { "family": "Song", "given": "Jialin" } }, { "id": "Tokpanov-Y-S", "name": { "family": "Tokpanov", "given": "Yury S." } }, { "id": "Chen-Yuxin", "name": { "family": "Chen", "given": "Yuxin" } }, { "id": "Fleischman-D", "name": { "family": "Fleischman", "given": "Dagny" } }, { "id": "Fountaine-K-T", "name": { "family": "Fountaine", "given": "Kate T." }, "orcid": "0000-0002-0414-8227" }, { "id": "Atwater-H-A", "name": { "family": "Atwater", "given": "Harry A." }, "orcid": "0000-0001-9435-0201" }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Optimizing Photonic Nanostructures via Multi-fidelity Gaussian Processes", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 1811.07707.pdf
", "abstract": "We apply numerical methods in combination with finite-difference-time-domain (FDTD) simulations to optimize transmission properties of plasmonic mirror color filters using a multi-objective figure of merit over a five-dimensional parameter space by utilizing novel multi-fidelity Gaussian processes approach. We compare these results with conventional derivative-free global search algorithms, such as (single-fidelity) Gaussian Processes optimization scheme, and Particle Swarm Optimization---a commonly used method in nanophotonics community, which is implemented in Lumerical commercial photonics software. We demonstrate the performance of various numerical optimization approaches on several pre-collected real-world datasets and show that by properly trading off expensive information sources with cheap simulations, one can more effectively optimize the transmission properties with a fixed budget.", "date": "2019-02-05", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20190205-101105728", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20190205-101105728", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.1811.07707", "primary_object": { "basename": "1811.07707.pdf", "url": "https://authors.library.caltech.edu/records/amrte-t0y10/files/1811.07707.pdf" }, "resource_type": "monograph", "pub_year": "2019", "author_list": "Song, Jialin; Tokpanov, Yury S.; et el." }, { "id": "https://authors.library.caltech.edu/records/c1zrc-dn749", "eprint_id": 90356, "eprint_status": "archive", "datestamp": "2023-08-19 07:52:33", "lastmod": "2023-10-18 23:23:37", "type": "monograph", "metadata_visibility": "show", "creators": { "items": [ { "id": "Zheng-Stephan", "name": { "family": "Zheng", "given": "Stephan" } }, { "id": "Yu-Rose", "name": { "family": "Yu", "given": "Rose" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" } ] }, "title": "Multi-resolution Tensor Learning for Large-Scale Spatial Data", "ispublished": "unpub", "full_text_status": "public", "note": "This result is supported in part by NSF #1564330, NSF #1637598, and gifts from Bloomberg and Northrop Grumman.\n\nSubmitted - 1802.06825.pdf
", "abstract": "High-dimensional tensor models are notoriously computationally expensive to train. We present a meta-learning algorithm, MMT, that can significantly speed up the process for spatial tensor models. MMT leverages the property that spatial data can be viewed at multiple resolutions, which are related by coarsening and finegraining from one resolution to another. Using this property, MMT learns a tensor model by starting from a coarse resolution and iteratively increasing the model complexity. In order to not \"over-train\" on coarse resolution models, we investigate an information-theoretic fine-graining criterion to decide when to transition into higher-resolution models. We provide both theoretical and empirical evidence for the advantages of this approach. When applied to two real-world large-scale spatial datasets for basketball player and animal behavior modeling, our approach demonstrate 3 key benefits: 1) it efficiently captures higher-order interactions (i.e., tensor latent factors), 2) it is orders of magnitude faster than fixed resolution learning and scales to very fine-grained spatial resolutions, and 3) it reliably yields accurate and interpretable models.", "date": "2018-10-23", "date_type": "published", "id_number": "CaltechAUTHORS:20181023-101356776", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20181023-101356776", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NSF", "grant_number": "CCF-156433" }, { "agency": "NSF", "grant_number": "CCF-1637598" }, { "agency": "Bloomberg" }, { "agency": "Northrop Grumman Corporation" } ] }, "doi": "10.48550/arXiv.1802.06825", "primary_object": { "basename": "1802.06825.pdf", "url": "https://authors.library.caltech.edu/records/c1zrc-dn749/files/1802.06825.pdf" }, "resource_type": "monograph", "pub_year": "2018", "author_list": "Zheng, Stephan; Yu, Rose; et el." } ]