| name | neural-manifolds-crystallized-embeddings |
| description | Neural manifolds as crystallized embeddings: a synthesis of free energy principle, generalized synchronization, and Hebbian plasticity. Proposes that neural manifolds emerge developmentally through three interacting processes: dynamical contraction (free energy minimization), generalized synchronization (reservoir computing embedding), and correlation-based Hebbian plasticity that crystallizes embedded manifolds into recurrent connectivity. Use when studying neural manifold formation, head-direction cells, grid cells, free energy principle in neural coding, Hebbian learning of attractor networks, or developmental emergence of neural representations. arXiv: 2605.04200
|
Neural Manifolds as Crystallized Embeddings
Paper: Vikas N. O'Reilly-Shah (2026). "Neural Manifolds as Crystallized Embeddings: A Synthesis of the Free Energy Principle, Generalized Synchronization, and Hebbian Plasticity"
arXiv: 2605.04200
Categories: q-bio.NC
Core Thesis
Mature neural manifolds (head-direction, grid-cell, stimulus-driven visual manifolds) are not genetically prespecified templates, but developmental products of three interacting processes:
- Dynamical contraction (free energy principle)
- Generalized synchronization (reservoir computing embedding theorems)
- Correlation-based Hebbian plasticity (crystallization into recurrent connectivity)
Three-Process Framework
1. Dynamical Contraction
- Free energy minimization drives the system toward low-dimensional manifolds
- Neural dynamics contract onto embedded submanifolds in state space
- Provides the geometric scaffold for representation
2. Generalized Synchronization
- Sensory-driven synchronization generates correlations across neural populations
- Reservoir computing embedding theorems ensure faithful representation
- Input statistics determine the geometry of the synchronization manifold
3. Hebbian Crystallization
- Hebbian plasticity acts on correlations from synchronization
- Crystallizes the embedded manifold into recurrent connectivity
- Yields autonomous continuous attractor network (when fixed point exists)
Key Synthesis
Links three theoretical frameworks:
- Free Energy Principle: explains why dynamics contract onto manifolds
- Reservoir Computing Embedding Theorems: guarantees representational capacity
- Contraction Theory of Hebbian Networks: explains developmental crystallization
Testable Predictions
- Dimensional thresholds: Specific thresholds for topological recovery of manifolds
- Developmental sensitivity: Critical periods for plasticity-dependent manifold formation
- Input statistics dependence: Attractor geometry depends on sensory input statistics
- Fixed point existence: Central open problem — whether Hebbian fixed point preserves embedding quality
Applications
- Head-direction cell system development
- Grid cell formation mechanisms
- Stimulus-driven visual manifold emergence
- Understanding developmental disorders of neural representation
- Brain-computer interface design (leveraging natural manifold structure)
Central Open Problem
Whether the Hebbian fixed point exists and preserves the embedding quality of the synchronization manifold. This determines whether the three-process synthesis fully accounts for observed neural manifold properties.
Activation Keywords
- neural manifold, crystallized embedding, free energy principle
- generalized synchronization, Hebbian plasticity, attractor network
- head-direction cells, grid cells, developmental neuroscience
- reservoir computing, contraction theory, continuous attractor
