Skip to main content

chaotic-regularization-recurrent-networks

Link microscopic chaos in recurrent neural networks to macroscopic geometry of neural representations using kernel methods and dynamical mean-field theory. Chaotic dynamics act as intrinsic regularizer enhancing generalization while preserving expressivity.

Zur Installation springen

Quellinformationen

Repository
hiyenwong/ai_collection
Letzte Quellaktivität
4. Juni 2026 um 13:32
Erkannte Sprache von SKILL.md
Englisch
Sterne
2
Forks
0

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

SKILL.md wird angezeigt

SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
chaotic-regularization-recurrent-networks
description
Link microscopic chaos in recurrent neural networks to macroscopic geometry of neural representations using kernel methods and dynamical mean-field theory. Chaotic dynamics act as intrinsic regularizer enhancing generalization while preserving expressivity.
keywords
["recurrent neural networks","chaos","neural representations","kernel methods","dynamical mean-field theory","cortical circuits","population codes","power-law spectral signatures","regularization"]
authors
["Jan Bauer","Christian Keup","Jonathan Kadmon","Moritz Helias"]
arxiv_id
2606.04426
date_added
2026-06-04T00:00:00.000Z
paper_url
https://arxiv.org/abs/2606.04426
pdf_url
https://arxiv.org/pdf/2606.04426
doi
https://doi.org/10.48550/arXiv.2606.04426
subjects
["q-bio.NC","cond-mat.dis-nn"]
# Chaotic Regularization in Recurrent Neural Networks ## Overview Cortical circuits operate in intrinsic chaos regimes, yet population codes vary smoothly with stimuli, forming coherent representational manifolds. This paper develops a theoretical framework linking microscopic chaos to macroscopic representation geometry, explaining how chaotic spiking networks sustain smooth, differentiable population codes. ## Core Methodology ### 1. Theoretical Framework - **Kernel Methods**: Combine kernel methods with dynamical mean-field theory - **Local vs Global Smoothness**: Chaotic dynamics induce: - **Local roughness**: Sharp distortions at small scales - **Global smoothness**: Preserved across larger stimulus variations - **Intrinsic Regularization**: Structural property acts as regularizer enhancing generalization while maintaining expressivity ### 2. Power-Law Spectral Signatures - Chaotic networks naturally produce power-law spectral signatures - Closely matches experimental observations in cortical recordings - Links network dynamics to recorded neural activity ### 3. Computational Structure - Establishes connection between: - Network dynamics - Computational structure - Recorded neural activity ## Key Insights ### Chaos as Benefit - **Challenge**: Tiny input changes → divergent neural responses - **Solution**: Chaos provides intrinsic regularization - **Result**: Smooth population codes emerge from chaotic dynamics ### Representation Geometry - Local roughness improves generalization - Global smoothness maintains expressivity - Power-law spectra match cortical observations ## Technical Implementation ### Dynamical Mean-Field Theory - Analyze chaotic dynamics in recurrent networks - Connect microscopic chaos to macroscopic representations - Derive spectral properties matching experiments ### Kernel Methods - Apply kernel methods to neural representations - Characterize local vs global smoothness - Quantify regularization effects ## Applications ### Brain Modeling - Explain chaotic spiking networks sustaining smooth codes - Model cortical circuit dynamics - Predict population code geometry ### AI/ML - Design chaotic regularization strategies - Balance expressivity vs generalization - Optimize recurrent network training ### Neuroscience Research - Interpret cortical power-law spectra - Connect dynamics to computational structure - Validate theoretical predictions experimentally ## Experimental Validation - Power-law signatures match cortical recordings - Population codes remain smooth despite chaos - Spectral properties align with observations ## Implementation Notes ### When to Use - Modeling cortical circuits with chaotic dynamics - Explaining smooth population codes from chaos - Designing regularization from network dynamics - Interpreting cortical spectral signatures ### Key Parameters - Chaotic regime strength - Kernel method selection - Mean-field approximations - Spectral signature validation ### Complementary Methods - Dynamical systems theory - Statistical mechanics - Kernel learning - Information geometry ## References - arXiv:2606.04426 (original paper) - Dynamical mean-field theory literature - Kernel methods in neural networks - Cortical dynamics experimental studies ## See Also - [[neural-critical-dynamics-theory]] - Critical dynamics in neural networks - [[chaos-freezing-without-plasticity]] - Chaos stabilization methods - [[efficient-coding-criticality-sloppiness]] - Efficient coding under constraints - [[representation-geometry-neural-networks]] - Representation geometry analysis
Auf GitHub ansehen