Explore a comprehensive lecture on advanced recurrent neural network models in cognitive neuroscience. Delve into Dr. Guangyu Robert Yang's insights on using machine learning-trained RNNs to explain complex behavior and neural activity patterns. Discover how these models generate rapid mechanistic hypotheses for cognitive computations and combine biological knowledge with computational goals. Examine the challenges faced in early works and learn about recent advancements in building next-generation RNN models. Investigate topics such as multi-task learning, continual learning, short-term memory mechanisms, and the development of multi-area models. Gain valuable perspectives on the future of artificial neural networks in understanding brains and minds.
Next-Generation Recurrent Network Models for Cognitive Neuroscience
Overview
Syllabus
Next Generation Recurrent Neural Network Models for Cognitive Neuroscience
How can artificial neural networks help us understand brains and minds?
Recurrent Neural Networks for cognitive neuroscience
How does the same circuit perform many tasks?
A single network is trained to perform many cognitive tasks
Clusters are causal to good performance
Continual learning of many cognitive tasks
Short-term memory Persistent activity, synaptic plasticity, or both?
Neural networks with short-term plasticity for short-term memory tasks
More persistent activity in lasks that require more manipulation
Opinion The bottleneck for multi-area models is not engineering
Discussion How do we build next-gen ANN models with style?
Taught by
MITCBMM
