Halassa Lab

Project 3

Explain the computational advantage of adding a thalamic-like architecture to recurrent networks:

Sub-project III-I We develop recurrent neural networks with a thalamus-like component and synaptic plasticity rules to model the thalamocortical interactions in cognitive flexibility. We find that the MD component is able to extract context information by integrating context-relevant traces over trials and to suppress context-irrelevant neurons in PFC. Incorporating the MD disjoints the contextual representations and enables efficient population coding in PFC, which shows the computational advantages in context switch and continual learning. (Wei-Long in collaboration with Prof. Robert Yang) 

Sub-project III-II We are developing a series of reservoir networks that communicate through thalamic intermediaries to solve probabilistic learning tasks. We find a role for MD in enhancing cognitive flexibility by responding to individual cortices, and also in mediating cortical interactions through a trans-thalamic route. Predictions for the model are compared with human fMRI data from subjects performing the same tasks. (Ali/Sabrina in collaboration with Prof. Burkhard Pledger and Dr. Bin Wang).

Sub-project III-III We further propose that thalamocortical-basal ganglia interaction serves as a system level solution for flexible and generalizable credit assignment. Specifically, we propose that basal ganglia guides the thalamocortical plasticity in two time scales to enable meta learning–the fast plasticity allows flexible contextual association while the slow plasticity develops a cortical representation that can generalize across context. Furthermore, our theory indicates an advantage of thalamocortical architectures in continual learning and working memory tasks. (Brabeeba)

Address

Department of Brain & Cognitive Sciences
43 Vassar St
Cambridge, MA 02139

Accessibility

MIT is committed to providing an environment that is accessible to individuals with disabilities.

You Are Welcome Here

Halassa Lab is committed to creating a diverse environment. All qualified applicants will receive consideration for employment without regard to race, color, religion, gender, gender identity or expression, sexual orientation, national origin, genetics, disability, age, or veteran status.