Search Papers | Poster Sessions | All Posters
Poster A27 in Poster Session A - Tuesday, August 6, 2024, 4:15 – 6:15 pm, Johnson Ice Rink
Inverse stochastic learning
Momchil Tomov1,2; 1Harvard University, 2Motional AD LLC
Rational process models posit that the brain learns the hidden structure of the world by approximating Bayesian inference using Monte Carlo sampling. The stochasticity of such inference algorithms makes it challenging to study the neural basis of the learning process, since it is difficult for the experimenter to know what the subject has inferred at any point in time. Here we tackle this inverse learning problem within the framework of inverse rational control using a simple particle filtering scheme. We evaluate our method on synthetic data and show that it uncovers the hidden states inferred by a subject on a trial-by-trial basis more accurately than a generative approach that only simulates the learning process. We then discuss how this method could be applied to a wide range of topics in cognitive computational neuroscience.
Keywords: structure learning Bayesian inference inverse rational control MCMC