[ { "id": "authors:m2n5a-kd302", "collection": "authors", "collection_id": "m2n5a-kd302", "cite_using_url": "https://resolver.caltech.edu/CaltechAUTHORS:20210920-171551019", "type": "book_section", "title": "Fine-Grained System Identification of Nonlinear Neural Circuits", "book_title": "Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining", "author": [ { "family_name": "Bagherian", "given_name": "Dawna", "clpid": "Bagherian-Dawna" }, { "family_name": "Gornet", "given_name": "James", "clpid": "Gornet-James" }, { "family_name": "Bernstein", "given_name": "Jeremy", "orcid": "0000-0001-9110-7476", "clpid": "Bernstein-Jeremy-D" }, { "family_name": "Ni", "given_name": "Yu-Li", "orcid": "0000-0003-1600-9854", "clpid": "Ni-Yu-Li" }, { "family_name": "Yue", "given_name": "Yisong", "orcid": "0000-0001-9127-1989", "clpid": "Yue-Yisong" }, { "family_name": "Meister", "given_name": "Markus", "orcid": "0000-0003-2136-6506", "clpid": "Meister-M" } ], "abstract": "We study the problem of sparse nonlinear model recovery of high dimensional compositional functions. Our study is motivated by emerging opportunities in neuroscience to recover fine-grained models of biological neural circuits using collected measurement data. Guided by available domain knowledge in neuroscience, we explore conditions under which one can recover the underlying biological circuit that generated the training data. Our results suggest insights of both theoretical and practical interests. Most notably, we find that a sign constraint on the weights is a necessary condition for system recovery, which we establish both theoretically with an identifiability guarantee and empirically on simulated biological circuits. We conclude with a case study on retinal ganglion cell circuits using data collected from mouse retina, showcasing the practical potential of this approach.", "doi": "10.1145/3447548.3467402", "isbn": "978-1-4503-8332-5", "publisher": "Association for Computing Machinery", "place_of_publication": "New York, NY", "publication_date": "2021-08-14", "pages": "14-24" }, { "id": "authors:38ne8-4rx67", "collection": "authors", "collection_id": "38ne8-4rx67", "cite_using_url": "https://resolver.caltech.edu/CaltechAUTHORS:20201106-120208748", "type": "book_section", "title": "Learning compositional functions via multiplicative weight updates", "author": [ { "family_name": "Bernstein", "given_name": "Jeremy", "orcid": "0000-0001-9110-7476", "clpid": "Bernstein-J-D" }, { "family_name": "Zhao", "given_name": "Jiawei", "clpid": "Zhao-Jiawei" }, { "family_name": "Meister", "given_name": "Markus", "orcid": "0000-0003-2136-6506", "clpid": "Meister-M" }, { "family_name": "Liu", "given_name": "Ming-Yu", "orcid": "0000-0002-2951-2398", "clpid": "Liu-Ming-Yu" }, { "family_name": "Anandkumar", "given_name": "Anima", "orcid": "0000-0002-6974-6797", "clpid": "Anandkumar-A" }, { "family_name": "Yue", "given_name": "Yisong", "orcid": "0000-0001-9127-1989", "clpid": "Yue-Yisong" } ], "contributor": [ { "family_name": "Larochelle", "given_name": "H.", "clpid": "Larochelle-H" }, { "family_name": "Ranzato", "given_name": "M.", "clpid": "Ranzato-M" }, { "family_name": "Hadsell", "given_name": "R.", "clpid": "Hadsell-R" }, { "family_name": "Balcan", "given_name": "M. F.", "clpid": "Balcan-M-F" }, { "family_name": "Lin", "given_name": "H.", "clpid": "Lin-H" } ], "abstract": "Compositionality is a basic structural feature of both biological and artificial neural networks. Learning compositional functions via gradient descent incurs well known problems like vanishing and exploding gradients, making careful learning rate tuning essential for real-world applications. This paper proves that multiplicative weight updates satisfy a descent lemma tailored to compositional functions. Based on this lemma, we derive Madam\u2014a multiplicative version of the Adam optimiser\u2014and show that it can train state of the art neural network architectures without learning rate tuning. We further show that Madam is easily adapted to train natively compressed neural networks by representing their weights in a logarithmic number system. We conclude by drawing connections between multiplicative weight updates and recent findings about synapses in biology.", "doi": "10.48550/arXiv.2006.14560", "publisher": "Advances in Neural Information Processing Systems", "publication_date": "2020-12" }, { "id": "authors:knyj1-fks47", "collection": "authors", "collection_id": "knyj1-fks47", "cite_using_url": "https://resolver.caltech.edu/CaltechAUTHORS:20200526-140034764", "type": "book_section", "title": "Synthetic Examples Improve Generalization for Rare Classes", "book_title": "2020 IEEE Winter Conference on Applications of Computer Vision (WACV)", "author": [ { "family_name": "Beery", "given_name": "Sara", "orcid": "0000-0002-2544-1844", "clpid": "Beery-S" }, { "family_name": "Liu", "given_name": "Yang", "clpid": "Liu-Yang" }, { "family_name": "Morris", "given_name": "Dan", "clpid": "Morris-Dan" }, { "family_name": "Piavis", "given_name": "Jim", "clpid": "Piavis-J" }, { "family_name": "Kapoor", "given_name": "Ashish", "clpid": "Kapoor-Ashish" }, { "family_name": "Meister", "given_name": "Markus", "orcid": "0000-0003-2136-6506", "clpid": "Meister-M" }, { "family_name": "Joshi", "given_name": "Neel", "clpid": "Joshi-Neel" }, { "family_name": "Perona", "given_name": "Pietro", "orcid": "0000-0002-7583-5809", "clpid": "Perona-P" } ], "abstract": "The ability to detect and classify rare occurrences in images has important applications - for example, counting rare and endangered species when studying biodiversity, or detecting infrequent traffic scenarios that pose a danger to self-driving cars. Few-shot learning is an open problem: current computer vision systems struggle to categorize objects they have seen only rarely during training, and collecting a sufficient number of training examples of rare events is often challenging and expensive, and sometimes outright impossible. We explore in depth an approach to this problem: complementing the few available training images with ad-hoc simulated data.Our testbed is animal species classification, which has a real-world long-tailed distribution. We present two natural world simulators, and analyze the effect of different axes of variation in simulation, such as pose, lighting, model, and simulation method, and we prescribe best practices for efficiently incorporating simulated data for real-world performance gain. Our experiments reveal that synthetic data can considerably reduce error rates for classes that are rare, that as the amount of simulated data is increased, accuracy on the target class improves, and that high variation of simulated data provides maximum performance gain.", "doi": "10.1109/WACV45572.2020.9093570", "isbn": "978-1-7281-6553-0", "publisher": "IEEE", "place_of_publication": "Piscataway, NJ", "publication_date": "2020-03", "pages": "852-862" }, { "id": "authors:52v8v-xx364", "collection": "authors", "collection_id": "52v8v-xx364", "cite_using_url": "https://resolver.caltech.edu/CaltechAUTHORS:20221222-175402264", "type": "book_section", "title": "Refractoriness and Neural Precision", "book_title": "Advances in neural information processing systems 10 : proceedings of the 1997 conference", "author": [ { "family_name": "Berry", "given_name": "Michael, II", "clpid": "Berry-Michael-II" }, { "family_name": "Meister", "given_name": "Markus", "orcid": "0000-0003-2136-6506", "clpid": "Meister-M" } ], "contributor": [ { "family_name": "Jordan", "given_name": "Michael J.", "clpid": "Jordan-Michael-J" }, { "family_name": "Kearns", "given_name": "Michael J.", "clpid": "Kearns-Michael-J" }, { "family_name": "Solla", "given_name": "Sara A.", "clpid": "Solla-Sara-A" } ], "abstract": "The relationship between a neuron's refractory period and the precision of its response to identical stimuli was investigated. We constructed a model of a spiking neuron that combines probabilistic firing with a refractory period. For realistic refractoriness, the model closely reproduced both the average firing rate and the response precision of a retinal ganglion cell. The model is based on a \"free\" firing rate, which exists in the absence of refractoriness. This function may be a better description of a spiking neuron's response than the peri-stimulus time histogram.", "isbn": "9780262100762", "publisher": "MIT Press", "place_of_publication": "Cambridge, MA", "publication_date": "1997-12", "pages": "110-116" } ]