[ { "id": "https://authors.library.caltech.edu/records/kpk97-10j19", "eprint_id": 94185, "eprint_status": "archive", "datestamp": "2023-08-19 19:52:03", "lastmod": "2023-10-20 17:45:43", "type": "conference_item", "metadata_visibility": "show", "creators": { "items": [ { "id": "Zhou-Jiaji", "name": { "family": "Zhou", "given": "Jiaji" } }, { "id": "Ross-S", "name": { "family": "Ross", "given": "Stephane" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Day-Debadeepta", "name": { "family": "Dey", "given": "Debadeepta" } }, { "id": "Bagnell-J-A", "name": { "family": "Bagnell", "given": "J. Andrew" } } ] }, "title": "Knapsack Constrained Contextual Submodular List Prediction with Application to Multi-document Summarization", "ispublished": "unpub", "full_text_status": "public", "note": "\u00a9 2013 by the author(s).\n\nPresented at the International Conference on Machine Learning (ICML) workshop on Inferning: Interactions between Inference and Learning, Atlanta, Georgia, USA, 2013. \n\nThis research was supported by NSF NRI Purposeful Prediction and the Intel Science and Technology Center on Embedded Computing. We gratefully thank Martial Hebert for valuable discussions and Alex Kulesza for providing data and code.\n\n
Submitted - 1308.3541.pdf
", "abstract": "We study the problem of predicting a set or list of options under knapsack constraint. The quality of such lists are evaluated by a submodular reward function that measures both quality and diversity. Similar to DAgger (Ross et al., 2010), by a reduction to online learning, we show how to adapt two sequence prediction models to imitate greedy maximization under knapsack constraint problems: CONSEQOPT (Dey et al., 2012) and SCP (Ross et al., 2013). Experiments on extractive multi-document summarization show that our approach outperforms existing state-of-the-art methods.", "date": "2019-03-27", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20190327-085828098", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20190327-085828098", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NSF" }, { "agency": "Intel Science and Technology Center for Embedded Computing" } ] }, "doi": "10.48550/arXiv.1308.3541", "primary_object": { "basename": "1308.3541.pdf", "url": "https://authors.library.caltech.edu/records/kpk97-10j19/files/1308.3541.pdf" }, "resource_type": "conference_item", "pub_year": "2019", "author_list": "Zhou, Jiaji; Ross, Stephane; et el." }, { "id": "https://authors.library.caltech.edu/records/xv8hz-r6a57", "eprint_id": 94622, "eprint_status": "archive", "datestamp": "2023-08-19 08:23:43", "lastmod": "2023-10-20 18:07:11", "type": "conference_item", "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": "Generating Multi-Agent Trajectories using Programmatic Weak Supervision", "ispublished": "unpub", "full_text_status": "public", "keywords": "deep learning, generative models, imitation learning, hierarchical methods, data programming, weak supervision, spatiotemporal", "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\nCode is available at https://github.com/ezhan94/multiagent-programmatic-supervision\n\nPublished - 1803.07612.pdf
", "abstract": "We study the problem of training sequential generative models for capturing coordinated multi-agent trajectory behavior, such as offensive basketball gameplay. When modeling such settings, it is often beneficial to design hierarchical models that can capture long-term coordination using intermediate variables. Furthermore, these intermediate variables should capture interesting high-level behavioral semantics in an interpretable and manipulable way. We present a hierarchical framework that can effectively learn such sequential generative models. Our approach is inspired by recent work on leveraging programmatically produced weak labels, which we extend to the spatiotemporal regime. 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": "2018-03-20", "date_type": "published", "publisher": "arXiv", "id_number": "CaltechAUTHORS:20190410-120555166", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20190410-120555166", "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 Corporation" } ] }, "doi": "10.48550/arXiv.1803.07612", "primary_object": { "basename": "1803.07612.pdf", "url": "https://authors.library.caltech.edu/records/xv8hz-r6a57/files/1803.07612.pdf" }, "resource_type": "conference_item", "pub_year": "2018", "author_list": "Zhan, Eric; Zheng, Stephan; et el." }, { "id": "https://authors.library.caltech.edu/records/re6pt-nap97", "eprint_id": 87329, "eprint_status": "archive", "datestamp": "2023-08-19 06:25:08", "lastmod": "2023-10-18 21:04:31", "type": "conference_item", "metadata_visibility": "show", "creators": { "items": [ { "id": "Su-Shihan", "name": { "family": "Su", "given": "Shihan" } }, { "id": "Chen-Yuxin", "name": { "family": "Chen", "given": "Yuxin" } }, { "id": "Mac-Aodha-O", "name": { "family": "Mac Aodha", "given": "Oisin" }, "orcid": "0000-0002-5787-5073" }, { "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" } ] }, "title": "Interpretable Machine Teaching via Feature Feedback", "ispublished": "unpub", "full_text_status": "public", "note": "The authors thank Google for supporting the Visipedia project, and kind donations from Northrop Grumman, Bloomberg, and AWS Research Credits. Yuxin Chen was supported in part by a Swiss NSF Mobility Postdoctoral Fellowship.\n\nPublished - nips17-teaching_paper-5.pdf
", "abstract": "A student's ability to learn a new concept can be greatly improved by providing them with clear and easy to understand explanations from a knowledgeable teacher. However, many existing approaches for machine teaching only give a limited amount of feedback to the student. For example, in the case of learning visual categories, this feedback could be the class label of the object present in the image. Instead, we propose a teaching framework that includes both instance-level labels as well as explanations in the form of feature-level feedback to the human learners. For image categorization, our feature-level feedback consists of a highlighted part or region in an image that explains the class label. We perform experiments on real human participants and show that learners that are taught with feature-level feedback perform better at test time compared to existing methods.", "date": "2017-12", "date_type": "published", "publisher": "Caltech Library", "id_number": "CaltechAUTHORS:20180622-113758617", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20180622-113758617", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "Northrop Grumman Corporation" }, { "agency": "Bloomberg" }, { "agency": "Amazon Web Services" }, { "agency": "Swiss National Science Foundation (SNSF)" } ] }, "primary_object": { "basename": "nips17-teaching_paper-5.pdf", "url": "https://authors.library.caltech.edu/records/re6pt-nap97/files/nips17-teaching_paper-5.pdf" }, "resource_type": "conference_item", "pub_year": "2017", "author_list": "Su, Shihan; Chen, Yuxin; et el." }, { "id": "https://authors.library.caltech.edu/records/ehp8b-sy192", "eprint_id": 77821, "eprint_status": "archive", "datestamp": "2023-08-19 02:27:05", "lastmod": "2023-10-25 23:28:58", "type": "conference_item", "metadata_visibility": "show", "creators": { "items": [ { "id": "Eyjolfsdottir-Eyrun-Anna", "name": { "family": "Eyjolfsdottir", "given": "Eyrun" } }, { "id": "Branson-K", "name": { "family": "Branson", "given": "Kristin" }, "orcid": "0000-0002-5567-2512" }, { "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": "Learning recurrent representations for hierarchical behavior modeling", "ispublished": "unpub", "full_text_status": "public", "note": "Submitted - 1611.00094
", "abstract": "We propose a framework for detecting action patterns from motion sequences and modeling the sensory-motor relationship of animals, using a generative recurrent neural network. The network has a discriminative part (classifying actions) and a generative part (predicting motion), whose recurrent cells are laterally connected, allowing higher levels of the network to represent high level behavioral phenomena. We test our framework on two types of data, fruit fly behavior and online handwriting. Our results show that 1) taking advantage of unlabeled sequences, by predicting future motion, significantly improves action detection performance when training labels are scarce, 2) the network learns to represent high level phenomena such as writer identity and fly gender, without supervision, and 3) simulated motion trajectories, generated by treating motion prediction as input to the network, look realistic and may be used to qualitatively evaluate whether the model has learnt generative control rules.", "date": "2017-04", "date_type": "published", "id_number": "CaltechAUTHORS:20170530-090151819", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20170530-090151819", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.48550/arXiv.1611.00094", "primary_object": { "basename": "1611.00094", "url": "https://authors.library.caltech.edu/records/ehp8b-sy192/files/1611.00094" }, "resource_type": "conference_item", "pub_year": "2017", "author_list": "Eyjolfsdottir, Eyrun; Branson, Kristin; et el." }, { "id": "https://authors.library.caltech.edu/records/74xh3-ysx85", "eprint_id": 75181, "eprint_status": "archive", "datestamp": "2023-08-19 01:45:40", "lastmod": "2023-10-25 14:49:22", "type": "conference_item", "metadata_visibility": "show", "creators": { "items": [ { "id": "Le-Hoang-M", "name": { "family": "Le", "given": "Hoang M." } }, { "id": "Carr-P", "name": { "family": "Carr", "given": "Peter" } }, { "id": "Yue-Yisong", "name": { "family": "Yue", "given": "Yisong" }, "orcid": "0000-0001-9127-1989" }, { "id": "Lucey-P", "name": { "family": "Lucey", "given": "Patrick" } } ] }, "title": "Data-Driven Ghosting using Deep Imitation Learning", "ispublished": "unpub", "full_text_status": "public", "note": "Published - 1671-2.pdf
", "abstract": "Current state-of-the-art sports metrics such as \"Wins-above-Replacement\" in baseball, \"Expected Point Value\" in basketball, and \"Expected Goal Value\" in soccer and hockey are now commonplace in performance analysis. These measures have enhanced our ability to compare and value performance in sport. But they are inherently limited because they are tied to a discrete outcome of a specific event. With the widespread (and growing) availability of player and ball tracking data comes the potential to quantitatively analyze and compare fine-grain movement patterns. An excellent example of this was the \"ghosting\" system developed by the Toronto Raptors to analyze player decision-making in STATS SportVU tracking data. Specifically, the Raptors created software to predict what a defensive player should have done instead of what they actually did. Motivated by the original \"ghosting\" work, we showcase an automatic \"data-driven ghosting\" method using advanced machine learning methodologies called \"deep imitation learning\", applied to a season's worth of tracking data from a recent professional league in soccer. Our ghosting method, which avoids substantial manual human annotation, results in a data-driven system that allows us to answer the question \"how should this player or team have played in a given game situation compare to the league average?\". In addition, by \"fine-tuning\" our league average model to the tracking data from a particular team, our ghosting technique can estimate how each team might have approached the situation. Our method enables counterfactual analysis of effectiveness of defensive positioning as both a measurable and viewable quantity for the first time.", "date": "2017-03", "date_type": "published", "publisher": "Caltech Library", "id_number": "CaltechAUTHORS:20170316-121646643", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20170316-121646643", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "primary_object": { "basename": "1671-2.pdf", "url": "https://authors.library.caltech.edu/records/74xh3-ysx85/files/1671-2.pdf" }, "resource_type": "conference_item", "pub_year": "2017", "author_list": "Le, Hoang M.; Carr, Peter; et el." } ]