Alladi Ramakrishnan Hall

New Learning Principles Emerge From Novel Biomimetic Computational Primitives

Anand Pathak

Department of Psychological and Brain Sciences, Dartmouth College, Hanover, NH, USA

What are the basic building blocks of thought and action? We created biologically realistic local circuits based on detailed physiological observations. These circuits may serve as primitive computational units, being irreducible assemblies. These biomimetic circuits were then integrated into a model of cortical-striatal interactions for category learning. Our model was not trained on, but evaluated against, neurophysiological recordings from non-human primates (NHPs) performing the same task. Remarkably, both the model and NHPs exhibited similar learning patterns. The model's simulated neurophysiology closely resembled actual NHP brain activity, capturing learning-induced spiking changes and beta synchrony between the prefrontal cortex and striatum. Significantly, our model made novel predictions that were subsequently confirmed in the NHP brain, such as the identification of “bad idea” neurons signaling impending incorrect choices. This is a rare instance of an entirely biologically inspired model, untrained but validated by neurophysiological data, yielding novel properties that were empirically affirmed.