Task learning increases information redundancy of neural responses in macaque visual cortex.

2026-03-05

Science

AI Summary

This study investigates how the primate brain optimizes sensory representations to support decision-making during visual task learning, specifically addressing whether neural response redundancy increases or decreases with learning. Through longitudinal electrophysiological recordings in macaque area V4, combined with population neural response analyses and information-theoretic methods, the work provides the first empirical support in primates for the Bayesian inference hypothesis. The findings reveal that task learning significantly enhances neural response redundancy both across weeks of training and within single trials. Crucially, this increased redundancy does not compromise information content; instead, it augments the information-carrying capacity of individual neurons. These results uncover a novel mechanism by which learning improves population coding efficiency through the strategic increase of redundancy.

Technology Category

Cognitive Modeling & Cognitive Systems: Neural Spike CodingMachine Learning: Bayesian LearningHumans and AI: Brain-Sensing and Analysis

Application Category

Economics, Online Markets and Human Computation: Social networks and social learningWeb Mining and Content Analysis: Content-based information diffusionUser Modeling, Personalization and Recommendation: Practical large-scale studies of user experience

Abstract

How does the brain optimize sensory information for decision-making in new tasks? One hypothesis suggests that learning reduces redundancy in neural representations to improve efficiency, whereas another, based on Bayesian inference, predicts that learning increases redundancy by distributing information across neurons. We tested these hypotheses by tracking population responses in macaque cortical area V4 as monkeys learned visual discrimination tasks. We found strong support for the Bayesian predictions: Task learning increased redundancy in neural responses over weeks of training and within single trials. This redundancy did not reduce information but instead increased the information carried by individual neurons. These insights suggest that sensory processing in the brain reflects a generative rather than discriminative inference process.

Problem

• task learning
• neural redundancy
• visual cortex
• Bayesian inference
• sensory information

Innovation

• neural redundancy
• Bayesian inference
• task learning
• V4 cortex
• generative inference