Hebbian-Oscillatory Co-Learning

Researcher Hasi Hays has introduced a novel bio-inspired AI framework called Hebbian-Oscillatory Co-Learning (HOC-L), detailed in a recent arXiv preprint. This unified two-timescale dynamical system jointly optimizes structural plasticity and phase synchronization in sparse neural architectures. HOC-L integrates two cutting-edge approaches: the hyperbolic sparse geometry of Resonant Sparse Geometry Networks (RSGN), which uses Poincaré ball embeddings and Hebbian-driven dynamic sparsity, and the oscillator-based attention mechanism of Selective Synchronization Attention (SSA), which replaces traditional dot-product attention with Kuramoto-type phase-locking dynamics.

The core innovation is 'synchronization-gated plasticity.' Here, the macroscopic order parameter r(t) of the oscillator ensemble acts as a gate for Hebbian structural updates. This means the network's physical connectivity only consolidates when sufficient phase coherence among artificial neurons signals a meaningful computational pattern, mimicking biological learning principles. Hays provides rigorous mathematical proofs for the system's convergence to a stable equilibrium using a composite Lyapunov function and derives explicit timescale separation bounds.

Numerical simulations confirm the theoretical predictions, showing emergent cluster-aligned connectivity and monotonic Lyapunov decrease. Crucially, the resulting architecture preserves the sparsity of its parent frameworks, achieving computational complexity of O(n·k) where k is much smaller than n (k≪n). This represents a significant step toward creating more efficient, interpretable, and neurologically plausible AI systems that learn structure and function simultaneously, moving beyond static, densely connected neural networks.