Please join us for a UW Data Science Seminar featuring Kris Bouchard, an Assistant Adjunct Professor at UC Berkeley, on Thursday, March 13th from 4:30 to 5:20 p.m. PT.
The seminar will be held in Hitchcock Hall 132 – Campus Map.
“Feedback control is a normative theory of neural population dynamics”
Abstract: Brain computations emerge from collective dynamics of distinct neural populations. Behaviors including reaching and speech are explained by principles of feedback control. However, if feedback control explains neural population dynamics is unknown. We created dimensionality reduction methods that identify subspaces of neural population data that are most feed-forward controllable (FFC) vs. feedback controllable (FBC). We show that FBC and FFC subspaces diverge for dynamics generated by neuro-anatomical connectivity. In neural recordings from across the brain, we show that FBC subspaces were better decoders of external variables (e.g. reach velocity, visual stimuli, animal location). Compared to FFC subspaces, FBC subspaces emerged from collective interactions of a population of neurons with distinct activity profiles. Finally, in M1/S1, we revealed that FBC subspaces emphasize rotational dynamics due to enhanced system stability, while FFC subspaces emphasize scaling dynamics. These results demonstrate feedback controllability is a novel, normative theory of neural population dynamics, and connect distinct neuronal populations to differing regimes of emergent dynamics carrying out distinct computations.
Biography: Kris Bouchard is Assistant Adjunct Professor in the Helen Wills Neuroscience Institute & Redwood Center for Theoretical Neuroscience at UC Berkeley. He is PI of the Neural Systems and Machine Learning Lab (NSML) at UCB and Lead of the Computational Biosciences Group in the Scientific Data Division, LBNL. The NSML lab is interdisciplinary team that focuses on understanding how distributed neural circuits give rise to coordinated behaviors and perceptions. We take a multi-pronged approach to this problem by developing novel theoretical frameworks for neural circuit function, conducting in vivo neuroscience experiments, and developing state of the art machine learning tools to address diverse systems biology questions. On the neuroscience side, we investigate functional organization and dynamic coordination in the brain by combining in vivo multi-scale electrophysiology and optogenetics in rodents. This multi-modal, multi-scale approach provides the simultaneous breadth of coverage and spatio-temporal resolution required to determine neural computations at the speed of thought. On the computational side, we aim to reveal the processes that generate complex (neuro-) biological data by combining ideas from control theory, information theory, high-dimensional statistical learning, statistical mechanics, and modern deep learning to develop novel methods and apply them to general biological and neuroscience problems.
The 2024-2025 seminars will be held in person, and are free and open to the public.