459

Source document

[2026-03-21 16:37] xenonfun: ``` FedSym Port: symbiogenesis → HarmonicRust Context The Python symbiogenesis project (github.com/MonumentalSystems/symbiogenesis) implements a population-based neural architecture evolution framework. Neural network "Units" (MLPs) fuse into increasingly complex architectures via horizontal gene transfer — sequential (deeper) or parallel (wider). Kuramoto oscillators synchronize the population and gelation phase transitions signal emergent complexity. FedSym beat all tested competitors in benchmarks. Goal: Port the evolutionary framework core into harmonic-core as a symbiogenesis module. Faithful Euclidean MLP implementation first (for validation against Python), but trait-abstracted so HarmonicRust's manifold-aware blocks can replace the Unit type in a future iteration. This becomes the reference implementation once behavior and performance are proven. Decisions resolved: - Scope: Core evolutionary framework — population, fusion, training, Kuramoto, gelation, MI, LR genes. No CNN (Phase 11b), no LoRA (Phase A), no NLP tasks. Those are V2. - Reproducibility: Behavioral/qualitative equivalence. Same algorithm, same convergence patterns, gelation at similar points. Not bit-for-bit. - Architecture: Faithful Euclidean MLPs with standard Adam, trait-abstracted for future manifold swap. - Training: Hand-written MLP backprop in harmonic-core (matches existing codebase pattern). Behind feature flag symbiogenesis. - Tasks MVP: XOR + sine + small MNIST subset. Full suite in V2. - CLI: infer evolve subcommand (follows existing pattern of infer train, infer infer, etc.) - Reservoir eigenvalues: nalgebra for eigenvalue decomposition (add as optional dep behind feature flag). ``` [2026-03-21 16:40] xenonfun: ✻ Creating scaffold and config… (4m 27s · ↓ 7.0k tokens · thought for 5s) ⎿  ◼ Step 1-2: Scaffold + Config + Enums ◻ Step 3: Activation functions ◻ Step 4-6: Unit + Training + Adam ◻ Step 7: LR Schedule Gene ◻ Step 8: Fusion ◻ Step 9-12: Population + Kuramoto + Monitor + MI ◻ Step 13-14: Reservoir + Tasks ◻ Step 15-18: Simulation + CLI + Tests