Kameron Harris

Western Washington University
, Online - Zoom

Abstract

Computational neuroscience: studying how the brain and brain-inspired algorithms function

People have studied the brain for centuries, but only in modern times have we been able to image individual neurons and unravel their functions. This challenge is far from solved. Now we can measure the activity of thousands of neurons at once, as well as the network those neurons use to interact. Mathematics have played an important role in suggesting theories of neuron and network function, as well as providing important tools to analyze high-dimensional datasets. I will provide a brief overview of this rich interdisciplinary field, before diving into my current interest of understanding artificial neural networks.

The theory of artificial neural networks based on kernels, a function that performs dot products in a nonlinear function space, is advancing our understanding of these powerful learning algorithms. Computational neuroscience is beginning to appreciate the relevance of kernel theory to brain networks. I will outline my perspective on these ideas and present results that come from adding elements of biological realism to networks and studying their associated kernels. In particular, we will see how realistic tuning properties and varying thresholds affect the space of functions that can be learned.