Research

Learning & Memory

Reinforcement Learning

The main topic of my postdoctoral career is to investigate the frontal-striatal circuitry that underlies reinforcement learning. To achieve this goal, I am applying two cutting-edge techniques to monkeys. In project 1, I use high electrode count neurophysiology to record large-scale neural activity in the cortical and subcortical regions underlie reinforcement learning. In Project 2, I implement fiber photometry recording in monkeys to directly measure dopamine release during reinforcement learning.

  • Information flow among frontal-striatal circuits.

  • Neural basis of novelty seeking.

  • What's dopamine's role in instrumental learning?

  • Neural coding of gains & losses, and does dopamine promote negative reinforcement?

Working Hypothesis: dorsal/ventral frontal-striatal circuits

Dorsal/ventral frontal-striatal circuits

A: The dorsal circuit includes dorsal LPFC, inferior parietal cortex (7a), dorsal striatum, globus pallidus internal segment (GPi) and the lateral portion of the medial dorsal (MD) thalamus. B: The ventral circuit includes the ventral MPFC, OFC, ventral striatum, ventral pallidum (VP) and medial portion of the MD thalamus.

Information flow in frontal-striatal circuits

During learning, the goal-related information, including the identity and the value of the to-be-chosen image, is maintained in the ventral circuit. Once it is time to make a choice, the information flows into the dorsal circuit, and transfers from object to direction information, flows from the caudal to the rostral part of the LPFC, to direct a saccade to the goal, which is the to-be-chosen image in our tasks.

Behavioral Tasks

Reversal learning task

A: Block structure and single-trial behavior. Blocks of 80 trials are used. In the acquisition phase, monkeys learn which target has a higher reward probability. In the middle, the reward mapping switches across targets. B: The reward probability associated with image identity in What blocks, associated with image location in Where blocks.

Token reward task

A: Diagram of the structure of a trial. In each trial, the monkey needs to choose between two images. The change of token number reflects the outcome of the choice. B: Each block consists of four images, stochastic associated with reward magnitudes of +2, +1, −1, and −2.

Novelty seeking task

A: Structure of an individual trial. Successive frames illustrate the sequence of events. B: Each session began with presenting three novel images, randomly assigned a reward probability of 0.2, 0.5 or 0.8. At the beginning of each new set, one of the existing options was randomly replaced with a novel image.

Working Memory

The amount of information stored in working memory is limited but may be improved with practice. However, the basis of improved efficiency at the neural level is less known. Recordings with arrays of microelectrodes as monkeys perform working memory tasks allow us to investigate the neural basis of working memory and how it can be improved.

  • How does training improve working memory?

  • How working memory is maintained and manipulated?

  • Cortical-subcortical circuits underlie working memory.