A Neural Information Field Approach to Computational Cognition
Abstract
The two main research objectives for this project are to understand how local cortical circuits embody states underlying cognitive functions, and to build large-scale models that are able to simulate a variety of tasks and levels of detail. Over the last three years, we have: demonstrated the first large-scale, function neural simulation to use highly detailed single neuron models, allowing the simulated testing of drug effects on cognitive performance; demonstrated a scalable neural model of motor planning; developed a new perceptual decision making model; demonstrated adaptive motor control in a large-scale cognitive simulation with spiking neurons (Spaun); demonstrated simple instruction following in Spaun; shown the first human-scale concept representations in spiking networks; demonstrated how to learn those representations; optimized cognitive computations for the Nengo simulation environment; demonstrated transfer learning to replicate performance of children learning to count in a SPA model; proposed a new SPA model of cognitive load using the N-back task; developed anew model of the effects of distraction in working memory; shown a hippocampal model able to perform context sensitive sequence encoding and retrieval; proposed what is currently the best model of neural activity during context-based working memory retrieval in monkeys; developed new software infrastructure for large spiking neural models; developed specialized hardware implementations of the N-back task; and optimized large model simulations for CPUs.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 18, 2016
- Accession Number
- AD1029686
Entities
People
- Chris Eliasmith
- Dimitris A. Pinotsis
Organizations
- University of Waterloo