Grant: Canada Research Chairs

The purpose of this research program is to better understand biological cognition and use that understanding to both improve medical interventions and enhance artificial systems faced with cognitive challenges. The research approach that we use is based in building large-scale, biologically detailed models of functioning brains. We have developed both a mathematical framework (the Neural Engineering Framework) and a cognitive architecture built on this framework (the Semantic Pointer Architecture), which have been used to create the world’s largest functional brain model, Spaun. In the proposed research program, we extend and apply this work in a number of ways. One core focus is to build a more embodied large-scale brain model, that is, to embed large-scale brain simulations in sophisticated, photo-realistic world simulations. Spaun has a single arm and a single, unmoving eye. In order to transition to a real, physical environment, next generation large-scale brain models need to control complex bodies in complex environments, to allow them to be transitioned to a real, physical environment. We have demonstrated the ability to develop artificial brains in physical simulations and transfer that training to real world quadcopters (i.e., drones). We believe this same approach can be applied to more sophisticated cognitive agents. This next generation of large-scale brain model will be novel not only for its scale and embodiment, but because it will include methods we have recently developed that capture cognitive representations and processing that are continuous in both time and state, allowing better performance in the real, continuous world. We have shown that these methods reflect recent neuroscientific observations, and that they beat state-of-the-art results on a wide range of machine learning benchmarks (e.g., SpeechCommands 2). Building these advanced methods into new biologically detailed brain models at unprecedented scales will push the boundaries of what can be achieved in integrated, functional brain modeling. We expect the results of this research to advance not only our understanding of how to build artificial cognitive agents, but to also significantly improve our understanding of how the biological brain works. Doing so would open the door to more effective medical interventions for a plethora of brain related diseases, both physical (e.g., stroke and Parkinson’s) and psychiatric (e.g., schizophrenia and bipolar disorder).