Publications
X.Lin, Z.Zheng., J.Topping*, W.Ma, Comparing Machine And Human Learning In A Planning Task Of Intermediate Complexity(Proceedings of the Annual Meeting of the Cognitive Science Society, 2022)
Conferences
Learning How Humans Play Board Games with GPT-4IAR (FCAI AI Day 2023)
Comparing Machine And Human Learning In A Planning Task Of Intermediate Complexity (The Multi-disciplinary Conference on Reinforcement Learning and Decision Making, 2022; Proceedings of the Annual Meeting of the Cognitive Science Society, 2022; )
Dissection of inter-area interactions of motor circuits (Simons Collaboration for the Global Brain Annual Meeting, 2022; Society for Neuroscience Annual Meeting, 2022)
- Abstract: Motor behaviors arise from dynamic interactions of interconnected neural populations across
distributed brain areas. The underlying principles of information flow remain largely
unknown. Here, we investigate the functional roles of motor cortex and intralaminar thalamus
in driving specific subpopulations of the striatum - the input nucleus of the basal ganglia -
during movements. We recorded the activity of direct and indirect pathway medium spiny
neurons (dMSNs and iMSNs) in the striatum as mice performed a skilled motor task, by in
vivo two-photon calcium imaging through GRIN lens. Furthermore, using monosynaptic
pseudo-typed rabies virus we identified and imaged corticostriatal and thalamostriatal neurons
that specifically project to dMSNs and iMSNs, through a glass window or a GRIN lens,
respectively. MSNs showed a sustained population activity throughout movement duration
that peaked at movement offset. Whereas, their cortical and thalamic inputs showed
contrasting activity dynamics, with corticostriatal activity concentrated around movement
onset and offset and thalamostriatal activity engaged during movement execution. To explore
activity differences among dMSNs and iMSNs and their inputs, we developed Trial Ensemble
Attention network (TEA-net) – a recurrent neural network with attention trained on ensembles
of single-trial neuronal activity. This approach followed by clustering analysis identified
quintessential activity patterns that were distinct between dMSNs and iMSNs and between
cortical and thalamic neurons that specifically project onto them. The results provide insights
on the mechanisms of integration of distinct long-range inputs carrying diverse information
by MSN subpopulations.