Linear Analysis of RNN Dynamics
Description
Recurrent neural networks (RNNs) are a powerful model for neural and cognitive phenomena. However, interpreting these models can be a challenge. In this tutorial, we will discuss how dynamical systems theory provides some tools for understanding RNNs. In particular, we will focus on the theory and application of linearizing RNN dynamics around fixed points. We will then look at some computational tools that have been developed around this framework.
Zoom meeting: https://mit.zoom.us/j/91318757507?pwd=NUlkL1NZTEFKcnNBRlNxUVVwQ213UT09
Password: 171132
Speaker Bio
Eli Pollock is a fifth year BCS PhD student in the Jazayeri lab. His research focuses on using RNNs to model cognitive processes.
Additional Info
We are always looking for more speakers to present in the series! If you would like to present at a tutorial, please contact Jenelle Feather (jfeather@mit.edu) or Nhat Le (nmle@mit.edu)