AC3: Neural Networks for Cognition

Description

Course content: In this course, we will look at how neural networks can help us to understand cognition. In particular, we will show how to construct simulations of biologically realistic neural networks, and how to organize these networks to produce cognitive behaviour. Most importantly, these neural systems will be shown to be strikingly different from traditional (non-neural) cognitive models. Understanding what types of algorithms neurons are good at implementing will (hopefully) lead us to new insights into how the mind works.

Session 1: Representing and Transforming Information with Neurons
Session 2: Building Complex Systems with Neurons
Session 3: Online and Offline Learning
Session 4: Neurons, Symbols, and Semantics

Objectives

Participants will be able to construct novel neural networks using the Nengo software package. They will also know how the behaviour of these networks differs from the behaviour of systems constructed with traditional programming approaches. Participants will see examples of neural models of many different brain functions, including vision, motor control, working memory, cognitive control, and decision making.

Literature

- Eliasmith, C. How to build a brain: A neural architecture for biological cognition.
Oxford University Press, New York, NY, 2013. http://nengo.ca/build-a- brain
- Bekolay, Bergstra, Hunsberger, DeWolf, Stewart, Rasmussen, Choo, Voelker, and
Eliasmith. Nengo: a Python tool for building large-scale functional brain models.
Frontiers in Neuroinformatics, 2014. http://compneuro.uwaterloo.ca/publications/bekolay2014.html

Course location

G√ľnne

Course requirements

TBA

Instructor information.

Instructor
Terry Stewart

Vita

Terry Stewart is a post-doctoral research associate with the Centre for Theoretical Neuroscience at the University of Waterloo. His PhD is in Cognitive Science from Carleton University, and emphasized the complexities involved in forming scientifically useful computational models of cognitive systems. Since 2008 he has worked with Chris Eliasmith on the development of the Neural Engineering Framework, and is both one of the lead software architects of Nengo and one of the
primary researchers making use of Nengo to generate and evaluate detailed computational models of complex cognitive functions. This includes Spaun, the first (and so far only) large-scale spiking neuron model capable of performing multiple
cognitive tasks.

Website

http://terrystewart.ca