Hidden Markov Model analysis of structured dynamics and internal state transitions during visual category learning in biological neural systems

Categorization is a fundamental cognitive process that enables organisms to classify similar, but nonidentical, stimuli as equivalent. Unlike artificial neural networks, which require extended training data and computational resources, biological systems can form categories quickly and efficiently. We investigated visual categorization learning in domestic chicks (Gallus gallus domesticus) and zebra finches (Taeniopygia guttata), using a «bead floor» model. Birds were presented with a plastic floor with color beads glued on it and food pellets. Peck-by-peck data were video recorded and analyzed using Hidden Markov Model (HMM) to reveal structured learning dynamics and internal state transitions indicative of cognitive processing. We subsequently introduced a modified paradigm with monochromatic bead floors and implemented a sequential training design. Chicks successfully formed new categories, and the learning process was modulated by their prior experience and stimuli order. To investigate the role of hemispheric processing in categorization, we implemented monocular training and testing. Category memory was confined only to the trained hemisphere when tested 3 hours after the training. These findings suggest that category memory in birds is shaped by prior similar experience and is restricted to the neural circuits that processed visual input.