Dopaminergic-cholinergic neuromodulation for rapid and democratic cortex-wide learning
Lead Research Organisation:
University of Oxford
Department Name: Physiology Anatomy and Genetics
Abstract
Learning depends on behavioural feedback. In the brain this feedback is encoded by diffuse neuromodulators. These neuromodulators are believed to trigger learning by assigning credit to synapses throughout the cortex. However, recent theoretical findings show that such diffuse credit assignment leads to very slow learning, in sharp contrast with animal and human learning. This raises a fundamental question: how can the cortex learn efficiently using diffuse neuromodulation?\
Here I propose that neuromodulators rely on excitatory-inhibitory circuits to assign credit efficiently to synapses throughout the cortex. Building on my expertise in developing computational models of cortical learning we will develop an integrative synapse-to-behaviour computational model of neuromodulated cortex- wide credit assignment. First, we will show that dopaminergic neuromodulation of excitatory-inhibitory cortical cell-types enables rapid reward-based credit assignment. Next, leveraging on adaptive machine learning principles I propose that cholinergic neuromodulation cooperates with the dopaminergic system to facilitate and robustify learning through democratic modulation of excitatory-inhibitory cortical circuits. To ensure that the model remains biologically and functionally sound, it will be contrasted with recent experimental data in close collaboration with leading experimentalists and machine learning researchers.
Finally, this dopaminergic-cholinergic model will generate testable predictions to the following questions: (i) what is the role of excitatory-inhibitory cell-types during goal-driven learning? (ii) how specific must neuromodulation be for rapid and robust goal-driven learning? and (iii) why is neuromodulatory malfunction commonly associated with cognitive decline in dementia and aging?
Overall, the proposed integrative computational framework, will be critical for our understanding of cortex- wide behaviourally-relevant learning in both health and disease.
Here I propose that neuromodulators rely on excitatory-inhibitory circuits to assign credit efficiently to synapses throughout the cortex. Building on my expertise in developing computational models of cortical learning we will develop an integrative synapse-to-behaviour computational model of neuromodulated cortex- wide credit assignment. First, we will show that dopaminergic neuromodulation of excitatory-inhibitory cortical cell-types enables rapid reward-based credit assignment. Next, leveraging on adaptive machine learning principles I propose that cholinergic neuromodulation cooperates with the dopaminergic system to facilitate and robustify learning through democratic modulation of excitatory-inhibitory cortical circuits. To ensure that the model remains biologically and functionally sound, it will be contrasted with recent experimental data in close collaboration with leading experimentalists and machine learning researchers.
Finally, this dopaminergic-cholinergic model will generate testable predictions to the following questions: (i) what is the role of excitatory-inhibitory cell-types during goal-driven learning? (ii) how specific must neuromodulation be for rapid and robust goal-driven learning? and (iii) why is neuromodulatory malfunction commonly associated with cognitive decline in dementia and aging?
Overall, the proposed integrative computational framework, will be critical for our understanding of cortex- wide behaviourally-relevant learning in both health and disease.
