[ { "id": "https://authors.library.caltech.edu/records/0k7zx-tfz21", "eprint_id": 106486, "eprint_status": "archive", "datestamp": "2023-08-20 00:35:14", "lastmod": "2023-12-22 23:40:53", "type": "book_section", "metadata_visibility": "show", "creators": { "items": [ { "id": "Huang-Yujia", "name": { "family": "Huang", "given": "Yujia" }, "orcid": "0000-0001-7667-8342" }, { "id": "Gornet-James", "name": { "family": "Gornet", "given": "James" }, "orcid": "0000-0002-5431-7340" }, { "id": "Dai-Sihui", "name": { "family": "Dai", "given": "Sihui" } }, { "id": "Yu-Zhiding", "name": { "family": "Yu", "given": "Zhiding" } }, { "id": "Nguyen-Tan-M", "name": { "family": "Nguyen", "given": "Tan" } }, { "id": "Tsao-D-Y", "name": { "family": "Tsao", "given": "Doris Y." }, "orcid": "0000-0003-1083-1919" }, { "id": "Anandkumar-A", "name": { "family": "Anandkumar", "given": "Anima" } } ] }, "title": "Neural Networks with Recurrent Generative Feedback", "ispublished": "unpub", "full_text_status": "public", "note": "We thank Chaowei Xiao, Haotao Wang, Jean Kossaifi, Francisco Luongo for the valuable feedback. Y. Huang is supported by DARPA LwLL grants. J. Gornet is supported by supported by the NIH Predoctoral Training in Quantitative Neuroscience 1T32NS105595-01A1. D. Y. Tsao is supported by Howard Hughes Medical Institute and Tianqiao and Chrissy Chen Institute for Neuroscience. A. Anandkumar is supported in part by Bren endowed chair, DARPA LwLL grants, Tianqiao and\nChrissy Chen Institute for Neuroscience, Microsoft, Google, and Adobe faculty fellowships.\n\n
Published - NeurIPS-2020-neural-networks-with-recurrent-generative-feedback-Paper.pdf
Supplemental Material - NeurIPS-2020-neural-networks-with-recurrent-generative-feedback-Supplemental.pdf
", "abstract": "Neural networks are vulnerable to input perturbations such as additive noise and adversarial attacks. In contrast, human perception is much more robust to such perturbations. The Bayesian brain hypothesis states that human brains use an internal generative model to update the posterior beliefs of the sensory input. This mechanism can be interpreted as a form of self-consistency between the maximum a posteriori (MAP) estimation of an internal generative model and the external environment. Inspired by such hypothesis, we enforce self-consistency in neural networks by incorporating generative recurrent feedback. We instantiate this design\non convolutional neural networks (CNNs). The proposed framework, termed Convolutional Neural Networks with Feedback (CNN-F), introduces a generative feedback with latent variables to existing CNN architectures, where consistent predictions are made through alternating MAP inference under a Bayesian framework. In the experiments, CNN-F shows considerably improved adversarial robustness over conventional feedforward CNNs on standard benchmarks.", "date": "2020-12", "date_type": "published", "publisher": "Advances in Neural Information Processing Systems", "id_number": "CaltechAUTHORS:20201106-120201944", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20201106-120201944", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "funders": { "items": [ { "agency": "NIH Predoctoral Fellowship", "grant_number": "1T32NS105595-01A1" }, { "agency": "Howard Hughes Medical Institute (HHMI)" }, { "agency": "Tianqiao and Chrissy Chen Institute for Neuroscience" }, { "agency": "Bren Professor of Computing and Mathematical Sciences" }, { "agency": "Defense Advanced Research Projects Agency (DARPA)" }, { "agency": "Learning with Less Labels (LwLL)" }, { "agency": "Microsoft Faculty Fellowship" }, { "agency": "Google Faculty Research Award" }, { "agency": "Adobe" } ] }, "local_group": { "items": [ { "id": "Tianqiao-and-Chrissy-Chen-Institute-for-Neuroscience" }, { "id": "Division-of-Biology-and-Biological-Engineering" } ] }, "contributors": { "items": [ { "id": "Larochelle-H", "name": { "family": "Larochelle", "given": "H." } }, { "id": "Ranzato-M", "name": { "family": "Ranzato", "given": "M." } }, { "id": "Hadsell-R", "name": { "family": "Hadsell", "given": "R." } }, { "id": "Balcan-M-F", "name": { "family": "Balcan", "given": "M. F." } }, { "id": "Lin-H", "name": { "family": "Lin", "given": "H." } } ] }, "doi": "10.48550/arXiv.2007.09200", "primary_object": { "basename": "NeurIPS-2020-neural-networks-with-recurrent-generative-feedback-Paper.pdf", "url": "https://authors.library.caltech.edu/records/0k7zx-tfz21/files/NeurIPS-2020-neural-networks-with-recurrent-generative-feedback-Paper.pdf" }, "related_objects": [ { "basename": "NeurIPS-2020-neural-networks-with-recurrent-generative-feedback-Supplemental.pdf", "url": "https://authors.library.caltech.edu/records/0k7zx-tfz21/files/NeurIPS-2020-neural-networks-with-recurrent-generative-feedback-Supplemental.pdf" } ], "resource_type": "book_section", "pub_year": "2020", "author_list": "Huang, Yujia; Gornet, James; et el." }, { "id": "https://authors.library.caltech.edu/records/fd3c9-3cd51", "eprint_id": 84193, "eprint_status": "archive", "datestamp": "2023-08-19 04:56:19", "lastmod": "2023-10-18 15:58:14", "type": "book_section", "metadata_visibility": "show", "creators": { "items": [ { "id": "Yoo-Sangjin", "name": { "family": "Yoo", "given": "Sangjin" }, "orcid": "0000-0002-0449-4242" }, { "id": "Sato-Tomokazu-F", "name": { "family": "Sato", "given": "Tomo" } }, { "id": "Tsao-D-Y", "name": { "family": "Tsao", "given": "Doris Y." }, "orcid": "0000-0003-1083-1919" }, { "id": "Shapiro-M-G", "name": { "family": "Shapiro", "given": "Mikhail" }, "orcid": "0000-0002-0291-4215" } ] }, "title": "Elucidating the biophysical mechanisms of ultrasonic neuromodulation", "ispublished": "unpub", "full_text_status": "restricted", "note": "\u00a9 2017 IEEE.", "abstract": "Ultrasonic neuromodulation is a promising technology in the field of basic and translational neuroscience because it can control neural activity within the skull without invasive surgery. In particular, low frequency ultrasound has been widely reported to elicit neural excitation and behavior in a number of animal models and humans. However, the cellular and molecular mechanisms of ultrasonic neuromodulation are not yet elucidated, impeding their use in neuroscience and potential clinical translation. To bridge this gap, we investigated the mechanisms of ultrasonic neuromodulation at the levels of single cells and intact organisms.", "date": "2017-09", "date_type": "published", "publisher": "IEEE", "place_of_pub": "Piscataway, NJ", "id_number": "CaltechAUTHORS:20180109-124523829", "isbn": "978-1-5386-3383-0", "book_title": "2017 IEEE International Ultrasonics Symposium (IUS)", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20180109-124523829", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "doi": "10.1109/ULTSYM.2017.8092954", "resource_type": "book_section", "pub_year": "2017", "author_list": "Yoo, Sangjin; Sato, Tomo; et el." }, { "id": "https://authors.library.caltech.edu/records/2en5n-p5b42", "eprint_id": 108819, "eprint_status": "archive", "datestamp": "2023-08-19 01:35:12", "lastmod": "2023-10-23 17:20:44", "type": "book_section", "metadata_visibility": "show", "creators": { "items": [ { "id": "Tsao-D-Y", "name": { "family": "Tsao", "given": "Doris Y." }, "orcid": "0000-0003-1083-1919" }, { "id": "Cadieu-Charles-F", "name": { "family": "Cadieu", "given": "Charles F." } }, { "id": "Livingstone-Margaret-S", "name": { "family": "Livingstone", "given": "Margaret S." } } ] }, "title": "Object Recognition: Physiological and Computational Insights", "ispublished": "unpub", "full_text_status": "public", "keywords": "face perception, face processing, object recognition, temporal lobe", "note": "\u00a9 2010 Oxford University Press.", "abstract": "Face perception is a microcosm of object recognition processes. The most difficult challenge in object recognition\u2014distinguishing among similar visual forms despite substantial changes in appearance arising from changes in position, illumination, occlusion, etc.\u2014is something we can do effortlessly for faces. Although face identification is often singled out as demanding particular sensitivity to differences between objects sharing a common basic configuration, in fact, such differences must be represented in the brain for both faces and nonface objects. This chapter argues that understanding face processing will illuminate the general problem of visual object recognition. It begins by discussing the functional architecture of the temporal lobe, with a special focus on the architecture of the system of face-selective areas in macaques and humans. It then discusses the physiology of cells in the temporal lobe, with a focus on the response properties of face-selective cells. Finally, it discusses different computational approaches to object recognition.", "date": "2010-02", "date_type": "published", "publisher": "Oxford University Press", "place_of_pub": "New York, NY", "pagerange": "471-499", "id_number": "CaltechAUTHORS:20210423-155556218", "isbn": "9780195326598", "book_title": "Primate Neuroethology", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20210423-155556218", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "contributors": { "items": [ { "id": "Platt-Michael-L", "name": { "family": "Platt", "given": "Michael L." } }, { "id": "Ghazanfar-Asif-A", "name": { "family": "Ghazanfar", "given": "Asif A." } } ] }, "doi": "10.1093/acprof:oso/9780195326598.003.0024", "resource_type": "book_section", "pub_year": "2010", "author_list": "Tsao, Doris Y.; Cadieu, Charles F.; et el." }, { "id": "https://authors.library.caltech.edu/records/kak79-req49", "eprint_id": 55199, "eprint_status": "archive", "datestamp": "2023-08-19 01:16:56", "lastmod": "2024-01-13 16:16:23", "type": "book_section", "metadata_visibility": "show", "creators": { "items": [ { "id": "Tsao-T-R", "name": { "family": "Tsao", "given": "Tien-Ren" } }, { "id": "Tsao-D-Y", "name": { "family": "Tsao", "given": "Doris Y." }, "orcid": "0000-0003-1083-1919" } ] }, "title": "Lie Group Model Neuromorphic Geometric Engine for Real-time Terrain Reconstruction from Stereoscopic Aerial Photos", "ispublished": "unpub", "full_text_status": "public", "keywords": "Lie group Model of Early vision, Stereoscopic imagery, 3-D model of Terrain, Real-time reconstruction, Analog Neuromorphic device", "note": "\u00a9 1997 SPIE. \n\nThe development of computational structure of neuromorphic geometric engine and its computer simulation and benchmark test was done by Dr. Thomas Tsao, funded by BMDO SBIR contract DASG6O-96-C-0058. The analog integrated circuit chip was designed and tested by Doris Tsao during her undergraduate study at Caltech, in relevant computational neural science courses.\n\nPublished - 1997-SPIE-3077-Tsao.pdf
", "abstract": "In the 1980's, neurobiologist suggested a simple mechanism in primate visual cortex for maintaining a stable and invariant representation of a moving object: The receptive field of visual neurons has real-time transforms in response to motion, to maintain a stable representation. When the visual stimulus is changed due to motion, the geometric transform of the stimulus triggers a dual transform of the receptive field. This dual transform in the receptive fields compensates geometric variation in the stimulus. This process can be modelled using a Lie group method. The massive array of affine parameter sensing circuits will function as a smart sensor tightly coupled to the passive imaging sensor (retina) . Neural geometric engine is a neuromorphic computing device simulating our Lie group model of spatial perception of primate's primal visual cortex. We have developed the computer simulation and experimented on realistic and synthetic image data, and performed a preliminary research of using analog VLSI technology for implementation of the neural geometric engine. We have benchmark tested on DMA's terrain data with their result and have built an analog integrated circuit to verify the computational structure of the engine. When fully implemented on ANALOG VLSI chip, we will be able to accurately reconstruct 3-D terrain surface in real-time from stereoscopic imagery.", "date": "1997-04-21", "date_type": "published", "publisher": "SPIE", "id_number": "CaltechAUTHORS:20150225-132447972", "official_url": "https://resolver.caltech.edu/CaltechAUTHORS:20150225-132447972", "rights": "No commercial reproduction, distribution, display or performance rights in this work are provided.", "contributors": { "items": [ { "id": "Rogers-S-K", "name": { "family": "Rogers", "given": "Steven K." } } ] }, "doi": "10.1117/12.271514", "primary_object": { "basename": "1997-SPIE-3077-Tsao.pdf", "url": "https://authors.library.caltech.edu/records/kak79-req49/files/1997-SPIE-3077-Tsao.pdf" }, "resource_type": "book_section", "pub_year": "1997", "author_list": "Tsao, Tien-Ren and Tsao, Doris Y." } ]