Эксперимент: Transfer-Spectrum Pseudo-Transition Scheduler (#496)

{ "worked": true, "confidence": 8, "verdict": "Built a compact transfer-spectrum scheduler prototype and an exact avoided-crossing verification in experiment.py. Across g=0.02, 0.05, 0.10, and 0.20, the measured gap/curvature response was centered at theta=0, the splitting matched 2g exactly, and xi matched the predicted base/(2g) within about 1.4%; the independently fitted memory decay also matched xi. In the small tanh-RNN adding-task comparison, the scheduler never triggered and baseline and idea were identical, so there is no observed practical training win; worked=true reflects the requested mechanism validation, not optimizer improvement.", "metrics": { "baseline": "RNN Adam: final loss 3.6649, best loss 1.6892, last-50-step mean 3.4575, final lr 0.01, detections 0", "idea": "Spectrum scheduler: final loss 3.6649, best loss 1.6892, last-50-step mean 3.4575, final lr 0.01, detections 0", "toy_predictions": "theta response/gap minimum 0.0 for all four couplings (predicted 0); splitting 0.04/0.10/0.20/0.40 versus predictions 2g; xi 24.997/9.992/4.983/2.466 versus 25/10/5/2.5; fitted xi slope versus 1/g 0.5005 (predicted 0.5); numerical memory xi 4.9833 versus predicted 4.9833" }, "how_to_run": "/home/maxwelhelp/main/bin/python3 experiment.py", "files": [ "experiment.py", "results.json", "run.log" ], "limitations": "Only an analytic two-mode toy transfer operator and one small one-seed RNN adding-task run were tested. The proposed scheduler was not stress-tested across learning rates or seeds, did not demonstrate damping or stability gains, and its entropy trigger remained inactive because the fitted RNN transfer spectrum was strongly single-mode." }