English
 
Help Privacy Policy Disclaimer
  Advanced SearchBrowse

Item

ITEM ACTIONSEXPORT

Released

Conference Paper

Why some simple probabilistic rules are difficult to learn: A hypothesis diffusion model

MPS-Authors
/persons/resource/persons293517

Teng,  T       
Department of Computational Neuroscience, Max Planck Institute for Biological Cybernetics, Max Planck Society;

Fulltext (restricted access)
There are currently no full texts shared for your IP range.
Fulltext (public)
There are no public fulltexts stored in PuRe
Supplementary Material (public)
There is no public supplementary material available
Citation

Chen, J., Teng, T., Shao, Y., & Zhang, H. (2024). Why some simple probabilistic rules are difficult to learn: A hypothesis diffusion model. In 2024 Conference on Cognitive Computational Neuroscience.


Cite as: https://hdl.handle.net/21.11116/0000-000F-C0E1-9
Abstract
People are known for their ability to learn probabilistic rewarding rules of the environment in both laboratory and real-world settings. However, in a series of experiments, we found that human participants failed to learn simple non-linear combinatory rules (e.g., the XOR rule of 11→1, 00→1, 01→0, 10→0 with a probability of 0.8), even after 320 trials, despite the small feature space for possible rules (i.e., including only two binary dimensions). This contrasted with the rapid learning of repetition or alternation rules in the same experiments. To explain why probabilistic XOR rules are difficult to learn, we propose a computational model that views rule learning as a progressively evolving hypothesis testing process. This hypothesis diffusion model assumes that (1) the weights assigned to different hypotheses diffuse across a network connecting hypotheses that can be transformed into each other through a single operation, and (2) the diffusion process is evidence-driven. The model successfully reproduces the behavioral patterns observed under each rule condition. Moreover, the model parameters estimated from a single rule condition can predict the observed differences in learning performance across different conditions.