Behavioural flexibility, the ability to adapt to change is a key component of natural intelligence. While artificial neural networks can now achieve human-like performance, adapting strategies or “changing the rules” remains a fundamental challenge. Although vital to survival, little is known about how humans and mammals, such as rodents, achieve this so effortlessly. In particular, it is not known which neurons in the cortical network are primarily responsible for rapid changes and what changes occur at the cellular level that facilitate such changes. Understanding how behavioural flexibility operates will involve a component-level to systems-level description that can be verified theoretically.
One of the key concepts that this research programme will explore is the computational power of cortical neurons and the extent to which this enhances the operation of the cortical circuitry. This complexity (multiple neurons with multiple compartments) affects both the operation of the network and the specific learning rules that can be implemented.
This collaboration between Dr. Panayiota Poirazi and well established behavioural neuroscience laboratories and theoretical groups in Berlin will address this problem. Dr. Poirazi will bring her long standing expertise in computational modelling of both complex, multi-compartmental neurons and simplified networks with various neuronal types. She will interact with the groups of Prof. Larkum and Prof. Schmitz in Berlin using ground-breaking experimental approaches to generate an abstract model of behavioural flexibility that is both biologically plausible and that can be implemented in a working artificial neural network.