State Space Neural Networks: A UWaterloo Invention

Thursday, September 5, 2024 11:30 am - 12:30 pm EDT (GMT -04:00)

 Join us for a seminar with Dr. Chris Eliasmith on neural networks.

Dr. Chris Eliasmith

Chris

Professor Chris Eliasmith is founding Director of the Centre for Theoretical Neuroscience at the University of Waterloo and holds the Canada Research Chair in Theoretical Neuroscience. He is jointly appointed to Systems Design Engineering and Philosophy, and cross-appointed to Computer Science. He has authored two books, 23 patents, and over 200 publications in philosophy, psychology, neuroscience, computer science, and engineering venues.

Abstract: State Space Neural Networks (SSNNs) have become an important new tool in the toolbox of deep learning researchers. Companies including Meta and Google now use them in large language models (LLMs) and other settings. In this talk I'll discuss the history of the development of SSNNs, their connection to biological brains, how they got their start at UW, and some of the nice properties of this kind of network. They are particularly good for time series problems, which can be traced to their provable optimality for time series data compression. I will conclude by providing pointers to some tools for incorporating SSNNs into your own work (e.g. the keras-lmu library).

This seminar is sponsored by: