Co-organizing minisymposium 261
Neuroscience 2019, the Society for Neuroscience (SfN) annual meeting, drew tens of thousands of brain scientists to McCormick Place in Chicago on October 19–23. Dr. Costilla Reyes, then a postdoctoral researcher in Earl Miller’s laboratory at MIT’s Picower Institute, proposed and co-organized minisymposium 261, “Artificial Intelligence and Neuroscience: From Neural Dynamics to Artificial Agents,” held on the morning of October 21 in room S406A and chaired by Maneesh Sahani. He had seen the call for minisymposia after the 2018 meeting and applied because he had seen few sessions at the intersection of AI and neuroscience, and he deliberately invited scientists who did not already collaborate; most were meeting one another in person at that session.
The program description says that data-analytic approaches have helped uncover dynamical structure in neural population activity, while artificial networks have given insight into representations and computations in the brain, and that the session would show how both fields can benefit from close interaction. The talks were by Lea Duncker on inferring variability from population spike trains, Joshua Glaser on state-space models for interacting neural populations, Hannah Choi on the dynamics of a mouse whole-brain connectome, Aran Nayebi on task-driven network models of higher visual cortex, and Andrea Banino on vector-based navigation in artificial agents.
Research presentation
Dr. Costilla Reyes gave talk 261.02, “Understanding neural dynamics under flexible visuomotor tasks with neural decoding and manifold learning,” on using machine learning to analyze the importance of different neural signals recorded in the cortex of experimental animals performing cognitive tasks. Neural decoding infers behavior or task variables from brain recordings, and manifold learning reveals low-dimensional structure in high-dimensional neural data, which is useful because the activity of many neurons often follows a much simpler pattern than the number of recorded signals suggests.
MIT’s Picower Institute reported that the session appeared to draw a large audience. Afterward Dr. Costilla Reyes thanked the speakers and proposed a joint review paper, for a Society for Neuroscience journal, based on the work presented. The session followed his Cambridge Science Festival talk and neural-decoding tutorial earlier that year, as part of his effort to connect artificial intelligence and neuroscience.