Recurrent Neural Networks in Brains and Machines

Neural networks are a powerful tool for modeling neural activity in the brain. In this talk, I will discuss how these models have helped in my own research and highlight recent work building neural networks that not only reproduce observed neural activity but also shed light on the underlying circuit mechanisms as they evolve a brain-like connectome through training. To more fully understand the brain, however, we need to take these modeling efforts beyond single circuits and tasks and consider the complex interactions between multiple systems: perception, cognition, action, memory, and many others. Achieving this goal requires models that bridge these systems, but this presents several challenges, including navigating the vast design space of potential models. To address this issue, I will describe ongoing work where we are developing a high-throughput pipeline to systematically build and evaluate recurrent neural network models on multiple datasets. By doing so, we hope to extract general design principles, or identify combinations of tasks, datasets, and models that are not currently well-explained, allowing us to continue to refine our understanding of both biological and artificial neural networks.

Learning Objectives:

1. Describe recurrent neural network (RNN) models.

2. Discuss ways in which RNNs are used to study the brain.

3. Evaluate the advantages and drawbacks of using RNNs as models for understanding the brain.