MRC Seminar: Active State Estimation, Learning, and Adaptive Control in Weakly Electric Fish
Friday, September 22, 2023
Active State Estimation, Learning, and Adaptive Control in Weakly Electric Fish
Noah J. Cowan
Johns Hopkins University
In this talk, I will present three interrelated studies on sensorimotor control in weakly electric fish.
(1) Active state estimation: The inescapable link between sensing and control generates a conflict between producing costly movements for gathering information (“explore”) versus using previously acquired information to achieve a goal (“exploit”). Solving the explore–exploit problem is computationally intractable for a wide variety of problems, requiring the use of heuristics. Here, we show data from phylogenetically diverse organisms, from amoebae to humans, that organisms solve the explore-vs.-exploit tradeoff using a mode-switching strategy, and we show that an implementation of this strategy significantly out-performs a traditional persistent excitation strategy. This work is in revision at Nature Machine Intelligence and a preprint is available: https://www.biorxiv.org/content/10.1101/2023.01.11.523566v3
(2) Learning to control destabilizing dynamics: Humans and other animals can readily learn to compensate for destabilizing dynamics, such as balancing an object or riding a bicycle. How does the nervous system learn to compensate for such dynamics, and what are the benefits of the newly learned control policies? To investigate these questions, we examined how the weakly electric glass knifefish, Eigenmannia virescens, retunes its control system in the face of destabilizing dynamics. Using a novel closed-loop system, we discovered that fish adapt their sensorimotor controllers as artificially destabilizing feedback is gradually introduced, and that the newly learned controller improves tracking performance and robustness. This work is in review and a preprint is available: https://www.biorxiv.org/content/10.1101/2023.01.27.525956v2
(3) Adaptive internal model principle (IMP) controller: It can be shown that perfect tracking of a periodic reference requires an internal model of the reference motion. It has been shown that humans and other animals indeed achieve nearly perfect tracking of periodic signals. We model this behavior using an adaptive internal model principle (IMP) controller and show that the model matches the time scales of adaptation in experiments with weakly electric fish. This work is in preparation for submission.
Noah J. Cowan received a B.S. degree from the Ohio State University, Columbus, in 1995, and M.S. and Ph.D. degrees from the University of Michigan, Ann Arbor, in 1997 and 2001 – all in electrical engineering. Following his Ph.D., he was a Postdoctoral Fellow in Integrative Biology at the University of California, Berkeley for 2 years. In 2003, he joined the mechanical engineering department at Johns Hopkins University, Baltimore, MD, where he is now tenured at the rank of Professor. Prof. Cowan has secondary appointments in Electrical and Computer Engineering and Computer Science. Prof. Cowan’s research interests include mechanics and multisensory control in animals and machines, and he has published scholarly articles in a diverse range of fields, from control systems and robotics to neuroscience and biomechanics. Prof. Cowan received the NSF PECASE award in 2010, the James S. McDonnell Foundation Scholar Award in Complex Systems in 2012, the William H. Huggins Award for excellence in teaching in 2004, and Johns Hopkins University Discovery Awards in 2015 and 2016.
Host: Ryan Sochol