Planning to plan: a Bayesian model for optimizing the depth of decision tree searchCogSci
July 27, 2021
I'm Ionatan, a third-year PhD student in neuroscience at NYU. I work with Wei Ji Ma, studying how people plan sequences of actions
in complex environments. This typically involves applying methods from artificial intelligence and reinforcement learning
to problems where the mental simulation of possible futures is intractable. Currently I am working on fitting process-level
models to large-scale naturalistic data, developing a framework based on Bayesian inference for optimizing planning depth in decision tree search,
and training deep neural networks to predict human gameplay in a challenging variant of tic-tac-toe. I'm also an NSF GRFP Fellow and president of the
Scientist Action and Advocacy Network. In my free time, I'm usually traveling or hiking for landscape photography, and I host an anime podcast.
Previously, I earned my B.A. in neuroscience, computer science, and mathematics at Macalester College. While I was an undergrad, I spent time in a few different labs working on projects throughout computational neuroscience. With David Heeger at NYU, I helped develop a new perceptual model for how humans compute optic flow fields, and with Srini Turaga at Janelia I worked on variational autoencoders for spike inference and connectivity. I did my honors thesis with Kendrick Kay at the University of Minnesota, on statistical analysis and visualization of high-resolution fMRI data. For my first rotation at NYU with Eero Simoncelli, I trained an optimized end-to-end image compression model based on the concept of divisive normalization in biological systems.