Hierarchical optimization predicts plasticity in the macaque inferior temporal cortex following object training.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The inferior temporal (IT) cortex is crucial for object recognition in the ventral visual stream.
• Understanding how the primate brain adapts its neural representations during object learning is key to deciphering visual processing plasticity.

Purpose of the Study:

• To investigate the neural consequences of object learning on the primate IT cortex.
• To model the observed changes in IT cortex using artificial neural network (ANN) simulations.
• To determine if plasticity in the ventral stream follows task optimization principles.

Main Methods:

• Electrophysiological recordings in male macaques performing an object discrimination task.
• Analysis of neural activity for object selectivity, linear separability, and representation invariance.
• Development and simulation of anatomically-mapped ANNs with various learning algorithms to model IT cortex plasticity.

Main Results:

• Task-trained monkeys exhibited increased object selectivity and enhanced linear separability in the IT cortex.
• Neural representations in trained monkeys became more object-invariant.
• Gradient-based learning algorithms in ANN models accurately replicated the observed changes in the IT cortex.
• Models predicted novel training-induced phenomena, including object-identity-independent changes.

Conclusions:

• Plasticity in the primate ventral stream, particularly the IT cortex, is driven by task optimization principles.
• Gradient descent provides a strong approximation for learning-induced changes in visual cortex.
• This convergence of empirical data and computational modeling allows for accurate predictions of visual plasticity and generalization.