Active-learning: Shaping rodent WM task performance

Whenever neuroscientists want to study the neural substrate of cognition, they first need to train animals to perform cognitive psychology tasks. Unfortunately, even tasks that a person could grasp after a 30 second verbal explanation, might take non-human animals several months to train. The goal of this project is to use deep reinforcement learning to model animal learning in a subset of memory tasks, and then conduct simulation experiments that might inform how to improve training time.

Goals

• use deep reinforcement learning to model rodent learning on working memory tasks
• use model to experiment with ways to improve training time.

WIP notes

• investigating basic implementation assumptions
  • stimulus coding
  • reward structure
  • task implementation
• working on curriculum experiments
  • pretrain on shorter delay
  • pretrain on less noisy stimuli

References

• PWM task