Bilinear Input-Conditioned Koopman Cell / report_bench_2026-09-04T173728.md
Mechanism confirmed, baseline not beaten
Стенд-проверка (stage-2) · промт оператора:
(универсальный)
Ответ агента:
{ "worked": false, "confidence": 9, "verdict": "Built and tested a rank-4 bilinear input-conditioned recurrent cell on the structurally matched actuated-pendulum dynamics track. It reduced mean MSE from 0.0020132 to 0.0015174 with 6/8 paired wins and confirmed an action-dependent trained-model signature, but the paired permutation p-value was 0.0991, so there is no significant win.", "metrics": { "baseline": "mean test MSE 0.0020132122, std 0.0006065339, lr=0.01, epochs=18, 8 seeds", "idea": "mean test MSE 0.0015174268, std 0.0001158377, rank=4, lr=0.01, epochs=18, 8 seeds", "paired_delta": "-0.0004957854", "permutation_p": "0.0991" }, "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_experiment.py", "files": [ "bench_experiment.py", "bench_report.json" ], "limitations": "Only the built-in dynamics track was tested. Dense bilinear factors, spectral constraints, multi-step free-running loss, MPC reward, and additional control environments were not evaluated.", "bench_report": { "bench_version": 1, "track": "dynamics", "model": "custom_bilinear_vs_additive", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.01, "epochs": 18 }, "sweep": [ { "cfg": { "lr": 0.001, "epochs": 18 }, "mean": 0.0190008804 }, { "cfg": { "lr": 0.003, "epochs": 18 }, "mean": 0.0066116862 }, { "cfg": { "lr": 0.01, "epochs": 18 }, "mean": 0.0020916119 } ], "full": { "mean": 0.0020132122, "std": 0.0006065339, "per_seed": [ 0.0018298299, 0.0025594952, 0.002444088, 0.0015330345, 0.0014931784, 0.0030939197, 0.0020086544, 0.0011434978 ], "n": 8 } }, "idea": { "mean": 0.0015174268, "std": 0.0001158377, "per_seed": [ 0.0015120441, 0.0015710213, 0.0013136059, 0.0014271569, 0.0016863476, 0.0014500043, 0.0016646207, 0.0015146137 ], "n": 8, "best_cfg": { "lr": 0.01, "epochs": 18 } }, "comparison": { "delta_mean": -0.0004957854, "idea_wins": 6, "n_pairs": 8, "per_seed_diffs": [ -0.0003177858, -0.0009884739, -0.0011304821, -0.0001058776, 0.0001931692, -0.0016439154, -0.0003440337, 0.000371116 ], "p_value": 0.0991, "verdict": "no significant win", "system_worked": false }, "architecture": "matched recurrent latent transition; only bilinear interaction differs", "parameter_note": "rank=4, hidden=48", "mechanism_signature": { "observed_action_flip_effect": 0.0343089055, "prediction_error": 0.0304896112, "ratio": 1.12526543, "confirmed": true } }, "system_verdict": "partial", "practical_verdict": "inconclusive", "mechanism_ok": 1, "system_judged": true }