Эксперимент: Spectral Memory-Lift Ensemble (#532)

{ "worked": true, "confidence": 9, "verdict": "Built a four-expert spectral memory lift with gated mixing and stable memory-rate parameterization in spectral_memory_lift.py. The kernel identity matched BD^(k-1)C to 5.6e-17, the measured stability transition matched the predicted gamma=1/rho(M0)=1.8814, and fitted exponential time constants matched theory across rho=0.5 to 0.99. In the fixed-seed delayed-memory regression, the ensemble achieved validation MSE 0.0724 versus 0.2954 for a single-exponential baseline, with bounded observed state/output norm 5.70; this is a promising toy win, not evidence for language-model superiority.", "metrics": { "baseline": "SingleExponential: validation MSE 0.2953793, train MSE 0.2901353, max output norm 1.0720", "idea": "SpectralMemoryLift with 4 experts: validation MSE 0.0723709, train MSE 0.0697623, max state norm 5.7043; kernel max error 5.55e-17; predicted/observed stability boundary gamma 1.88144/1.88144; fitted timescales matched theory to numerical precision" }, "how_to_run": "/home/maxwelhelp/main/bin/python3 spectral_memory_lift.py", "files": [ "spectral_memory_lift.py", "results.json" ], "limitations": "Only a small synthetic delayed-memory regression was tested, not enwik8, TinyStories, GRU or SSM equal-parameter language modeling, throughput, long-context perplexity, or optimizer spectral regularization/projection. The stability sweep uses exact eigendecomposition rather than the proposed approximate power-iteration penalty." }