Dataset for "Revealing how uncertainty computations drive hierarchical reasoning via CogLink Networks"
Description
This dataset contains experiments done in the paper "Revealing how uncertainty computations drive hierarchical reasoning via CogLink Networks". We run CogLink Network, its various perturbed version and other machine learning algorithms in various bandit tasks to understand the computational roles each neural mechanism serves in hierarchical reasoning under uncertainty. We show that these mechanistic features in CogLink Network enable efficient and flexible environmental exploration while replicating experimentally-observed neural signatures. Furthermore, it shows utility in computational psychiatry, by linking prefrontal dysfunction to idiosyncratic reasoning in schizophrenia.
Files
Steps to reproduce
Follow the instruction in the GitHub link. train.py reproduces all the experimental data while plot.py generates all the accompanied analysis and figures.
Institutions
- Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory