{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreihejlxfsf4vxu73pnnr3mghtqgqjp53mbzegmun7c7sn5tnxr6hau",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mh3zuxifga52"
  },
  "path": "/t/wave-coherence-scoring-phase-aware-alternative-to-cosine-similarity/173375#post_10",
  "publishedAt": "2026-03-15T11:08:22.000Z",
  "site": "https://discuss.huggingface.co",
  "tags": [
    "kerr-memory"
  ],
  "textContent": "Wave Memory: Persistent Experience for Kerr-ODE Models\n\nNew investigation results and a fourth repo.\n\n**The concept:** Model weights are the education — frozen, never change. A separate 1.5KB file stores accumulated harmonic band states — the experience. Each conversation shifts the Kerr-ODE’s starting position on the unit circle. Same model, different trajectory, different output. The model reads and writes memory in the same coordinate system it thinks in — no translation, no vector database, no RAG retrieval.\n\n**The mechanism:** During inference, the ODE final states feed an exponential moving average. Bands consistently active across tokens accumulate. Bands that spike once and fade contribute nothing. At conversation end, the accumulator merges into the persistent file. Next conversation starts from a different position because of what came before.\n\n**5 experiments, 4 passes, 1 honest null:**\n\nExperiment | Result\n---|---\nInjection sensitivity | Random noise at α=0.05 **improves perplexity by 8.8%** (stochastic resonance)\nAccumulation stability | Converges over 20 conversations, growth rate 280%→9.5%\nTopic separation (char-level) | **NULL** — captures corpus texture, not topic. Bounded by model capacity\nTopic separation (word-level) | **PARTIAL POSITIVE** — love→\"fair\",“give thee” vs war→\"dishonour\",“death” from same prompt\nMemory reset | Bit-identical to baseline after deletion\nAnomaly detection | Spike caught immediately before affecting output\n\nThe stochastic resonance finding is the unexpected one. Injecting random noise into ODE initial conditions makes the model generate _better_ , not worse. The nonlinear Kerr dynamics (self-phase modulation, cross-phase coupling) use the perturbation constructively. Standard transformers degrade with any perturbation — they have no mechanism to exploit noise. The Kerr-ODE does.\n\nThe topic separation scaled with tokenisation: character-level saw only texture (0.987 correlation between love and war memories). Word-level saw tone — same top bands but 2x energy difference, reordered peaks, and measurably different generation. BPE with a larger model is predicted to enable full semantic separation.\n\n**The safety model:** Delete the file → model returns to trained baseline (verified bit-identical). Inspect the file → harmonic census shows exactly what accumulated. The model weights never change during inference. Memory is experience, not education. Neither corrupts the other.\n\n**The repos:**\n\n  * kerr-memory — wave memory library, 920 lines Rust, zero dependencies (Apache 2.0)\n\n  * kerr-server — serves models with `--memory` flag (Apache 2.0)\n\n  * kerr-engine — trains models that produce the checkpoints (Apache 2.0)\n\n  * Wave Coherence — research framework, 70 defensive patents (MIT)\n\n\n\n\nFull pipeline works: train → serve → chat → accumulate → save → inspect. All Rust, all open source.",
  "title": "Wave-Coherence Scoring: Phase-Aware Alternative to Cosine Similarity"
}