Rapid context inference in a thalamocortical model using recurrent neural networks.

Area of Science:

• Computational Neuroscience
• Cognitive Neuroscience

Background:

• Cognitive flexibility is essential for context-appropriate behavior.
• Thalamocortical circuits, particularly prefrontal cortex (PFC) and mediodorsal thalamus (MD) interactions, are vital for temporal context inference.
• The precise neural mechanisms underlying context inference remain unclear.

Purpose of the Study:

• To propose and investigate a PFC-MD neural circuit model for rapid, online context inference.
• To elucidate the neural mechanisms enabling cognitive flexibility and continual learning.

Main Methods:

• Developed a computational model of PFC-MD interactions incorporating Hebbian plasticity.
• Implemented pre-synaptic traces, adaptive thresholding, and winner-take-all normalization in the model.
• Simulated sequential learning of cognitive tasks to evaluate model performance.

Main Results:

• The model MD thalamus successfully inferred temporal contexts from PFC inputs within a few trials.
• The PFC-MD model demonstrated continual learning by gating context-irrelevant PFC neurons.
• The model alleviated catastrophic forgetting and showed knowledge transfer across tasks.

Conclusions:

• Biological properties of thalamocortical circuits can support rapid context inference and continual learning.
• The proposed PFC-MD model offers insights into the neural basis of cognitive flexibility.
• Hebbian plasticity in thalamocortical circuits is a key mechanism for adaptive behavior.