Overview
Explore the concept of the prefrontal cortex as a meta-reinforcement learning system in this 55-minute lecture by Matthew Botvinick from DeepMind Technologies Limited and University College London. Delve into computational theories of the brain, covering topics such as recurrent neural networks, neuroscience and AI, and learning to learn. Examine the connections between Brendan Lake's work on Atari games and background knowledge, and investigate the differences between learning and inference. Analyze various neural network architectures, including multilayer perceptrons and recurrent neural networks, and their applications to bandit problems and Harlow's task. Discover insights from neuroscience research on hippocampal amnesia and the inferred value effect. Gain a comprehensive understanding of how these concepts contribute to a virtuous circle of collaboration between AI and neuroscience.
Syllabus
Introduction
Why is it relevant
Recurrent Neural Networks
Neuroscience and AI
Brendan Lake and Atari
Brendans Background Knowledge
Learning to Learn
Learning vs Inference
Multilayer Perceptron
Recurrent Neural Network
Bandit Problems
Harlows Task
Two Neuroscience
Hippocampal Amnesia
Neuroscience
Inferred Value Effect
Other Simulations
Summary
virtuous circle
collaborators
Taught by
Simons Institute