Dartmouth Events

Abstract: We simulate and analyze the behavior of a common excitatory-inhibitory circuit found in the superficial layers of the cortex. This circuit performs different computations depending on the strength of feedback from local inhibitory cells.
When feedback inhibition is strong, the circuit naturally carries out categorization. This behavior matches clustering and hierarchical clustering, two well-known methods in machine learning. When feedback inhibition is weak, the same circuit performs Principal Component Analysis (PCA). Both types of computation arise from the circuit’s same structure and different dynamics.

Based on the activity patterns observed in this brain circuit, we demonstrate a mathematical relationship between clustering algorithms and PCA. This connection is a rare example of how these two traditionally separate algorithms relate to each other within a unified framework.

Thesis Committee: Richard Granger (Chair), Peter Chin, Simon G. Shepherd, Lilianne R. Mujica-Parodi (Stony Brook University)