Inhibitory normalization of error signals improves learning in neural circuits

2026-03-18

AI Summary

This study investigates how normalization mediated by inhibitory neurons influences the learning capacity of neural circuits, with a particular focus on its role in error signal processing. To this end, we constructed artificial networks incorporating both excitatory and inhibitory neurons and evaluated learning performance on an image recognition task under varying illumination conditions, comparing scenarios where inhibitory normalization was applied only during forward inference versus both forward and backward passes. The results demonstrate that model performance significantly improves only when inhibitory normalization is implemented in both the forward and backward pathways; applying it solely during inference yields no benefit. This finding reveals for the first time that inhibitory normalization must act upon error signals to effectively facilitate learning, offering a novel perspective on the relationship between normalization mechanisms and learning in the brain.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Bio-inspired Learning
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Economics, Online Markets and Human Computation: Social networks and social learning
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Social Networks and Social Media: Influence propagation, information diffusion, and the prediction on networks

Abstract

Normalization is a critical operation in neural circuits. In the brain, there is evidence that normalization is implemented via inhibitory interneurons and allows neural populations to adjust to changes in the distribution of their inputs. In artificial neural networks (ANNs), normalization is used to improve learning in tasks that involve complex input distributions. However, it is unclear whether inhibition-mediated normalization in biological neural circuits also improves learning. Here, we explore this possibility using ANNs with separate excitatory and inhibitory populations trained on an image recognition task with variable luminosity. We find that inhibition-mediated normalization does not improve learning if normalization is applied only during inference. However, when this normalization is extended to include back-propagated errors, performance improves significantly. These results suggest that if inhibition-mediated normalization improves learning in the brain, it additionally requires the normalization of learning signals.

Problem

• inhibitory normalization
• error signals
• neural circuits
• learning improvement
• biological neural networks

Innovation

• inhibitory normalization
• error signal normalization
• excitatory-inhibitory networks
• learning improvement
• biological plausibility