I am a postdoc in Nao Uchida's Lab at the MCB dept.
I obtained my Ph.D at Harvard under Bernardo Sabatini in 2021.
I am broadly interested in how animals 'learn-to-learn'. Currently, I'm exploring how the brain toggles between stable and flexible representation in order to compute value optimally, and how this process might be impaired through aging. I use a combination of behavioral analysis, in vivo electrophysiology, fiber photometry, combined with RNN modelling.
We show that meta-learning in value computation emerges through a transition from plasticity-based value update to recurrent dynamics-based value update. We developed a new RNN implementation with an online learning rule (T-BPTT) which mimics mouse behavior and display meta-learning via a transition from plasticity to dynamics-based value update. This transition to dynamics-based value allow both RNNs and mice to infer value using structural knowledge.
In this perspective, we discuss recent works regarding the function of striatal indirect pathway. We focus on two ideas: dopamine dependent learning rule in the indirect pathway, and the action switching phenotype. We propose that these two facets allows the indirect pathway to avoid non-optimal actions while exploring alternative actions.
We dissected the functional impact of stimulating the striatal D2 indirect pathway, on behavior and neural dynamics in a downstream region (colliculus). We show that D2 pathway can both suppress and trigger movement on the contra/ipsi side respectively. This can be explained by paradoxical excitation of contralateral SC. We propose a general circuit mechanism for exploration mediated by basal ganglia-collicular circuit.
We show topographical loops starting from striatum to Pf, VM and SC in the mouse brain. Segregated pathways in medial vs ventrolateral striatum modulate turning/licking behavior respectively, confirming parallel organization of BG output pathways.
We show the application of tapered fibers in depth-selective photometry. We simultaneously recorded dopamine signals in dorsal/ventral striatum in the same mouse and show signals can diverge at specific behavioral events.