Evaluating Learning Rules in Visual Perceptual Learning Using Deep Neural Networks

This research project aims to explore and outline the implications of supervised and reinforcement learning dynamics in Deep Neural Networks (DNNs) with regard to Visual Perceptual Learning (VPL). DNNs, which are hierarchical computational models inspired by the biological brain, will be used to simulate various VPL tasks. The focus is on assessing how these distinct learning approaches—supervised, where networks are explicitly trained with correct responses, and reinforcement, where learning occurs via trial and error feedback—affect task performance and learning transferability. These effects will be evaluated by comparing the DNN outcomes with known human and animal perceptual learning characteristics The ultimate goal is to enhance our understanding of the underlying mechanisms of perceptual learning and to determine which learning paradigm most accurately replicates biological processes. This study promises to contribute significantly to our theoretical understanding of VPL and provide valuable insights into the design of more efficient, biologically-inspired artificial learning systems.

Faculty Supervisor:

Shahab Bakhtiari

Student:

Partner:

University of Tübingen

Discipline:

Life Sciences

Sector:

Life Sciences (not health); Artificial Intelligence; Other

University:

Université de Montréal

Program: