Examining Bio-Inspired Approaches for Continual Reinforcement Learning

• Author / Creator

  Mastikhina, Olya

• Despite the brain's inherent ability to continually learn, biological insights are rarely applied to continual reinforcement learning (RL).
  This thesis addresses this gap by examining four under-investigated biologically-inspired modifications within the context of continual RL: energy minimization, wire length constraints, sparse distributed memory multilayer perceptrons, and fuzzy tiling activations. We show that some of these modifications help increase plasticity and generalization as well as slightly decrease catastrophic forgetting. We additionally provide an analysis of the learned representations.

• Subjects / Keywords

  • reinforcement learning
  • continual learning
  • plasticity

• Graduation date

  Fall 2024

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/r3-0ny9-1h26

• License

  This thesis is made available by the University of Alberta Library with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.