Applying machine learning tools to examine the role of the hippocampus in reinforcement learning

Learning to optimize behaviour to maximize future reward is a central to a species' survival. Because the hippocampus is critical for memory recall and future planning, and capable of the phenomenon of replay (the spontaneous reactivation of neural sequences that represent past or future behaviour), it is believed to play a central role in reinforcement learning. In this project we will investigate the role of the hippocampus and neural replay in a behaving rat exploring an environment to optimize future reward, and examine two different types of reinforcement learning- model-free and model-based. We will rely on machine learning tools to decode neural activity during behaviour, and adapt these methods to decode awake replay.
There are three main aims to the project. The first is the development of machine learning tools that can determine the location of the animal based on hippocampal activity. This will then be applied to decoding the content of awake replay events, which presents a potential challenge in machine learning algorithms due to awake replay occurring on a time scale 10-20x faster than similar sequences driven by behavioural trajectories. Next, rats will be trained in a behavioural task, to find hidden reward zones in an open arena, which can be manipulated to examine whether animals learn fixed trajectories (model-free) or can adapt quickly to create a new trajectory to maximize rewards (model-based). Finally, using high-density, chronic neural recordings in behaving rodents, we will examine how replayed neural activity reflects the past and future behavior of the animal, specifically if it represents the most visited or highly rewarded trajectories or explores novel trajectories.

Supervision in this project will be provided by Dr Bendor and Dr Robinson with expertise in the neurobiological and computational methods required (Bendor lab- neurophysiological recordings in the rodent hippocampus; Robinson lab- advanced image processing/machine-learning.