Decomposing stimulus-specific sensory neural information via diffusion models

2025-05-16

AI Summary

In neural coding, decomposing mutual information into interpretable, stimulus- and feature-specific contributions has long been hindered by ill-posedness and the absence of axiomatic foundations. Method: We propose the first stimulus-specific information decomposition framework satisfying three axioms—additivity, positivity, and locality—thereby enabling semantic-level溯源 of neural information under high-dimensional natural stimuli. Our approach integrates axiomatic information-theoretic modeling with a diffusion-model-based mutual information estimator, overcoming classical estimation bottlenecks in information decomposition. Results: Validated on biophysically plausible visual neuron models, our framework precisely quantifies the independent contributions of individual stimuli and their features to neural encoding. It significantly enhances scalability, interpretability, and generalizability in neural representation analysis—enabling rigorous, axiomatically grounded dissection of how sensory information is partitioned across stimulus dimensions.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Information Theory
• Natural Language Processing: Information Extraction

Application Category

• Web Mining and Content Analysis: Content-based information diffusion
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

To understand sensory coding, we must ask not only how much information neurons encode, but also what that information is about. This requires decomposing mutual information into contributions from individual stimuli and stimulus features fundamentally ill-posed problem with infinitely many possible solutions. We address this by introducing three core axioms, additivity, positivity, and locality that any meaningful stimulus-wise decomposition should satisfy. We then derive a decomposition that meets all three criteria and remains tractable for high-dimensional stimuli. Our decomposition can be efficiently estimated using diffusion models, allowing for scaling up to complex, structured and naturalistic stimuli. Applied to a model of visual neurons, our method quantifies how specific stimuli and features contribute to encoded information. Our approach provides a scalable, interpretable tool for probing representations in both biological and artificial neural systems.

Problem

• Decompose neural information into stimulus-specific contributions
• Develop axioms for meaningful stimulus-wise decomposition
• Estimate decomposition using scalable diffusion models

Innovation

• Decompose neural info via diffusion models
• Introduce three core axioms for decomposition
• Scalable interpretable tool for neural systems