Semi-Passive Energy-Gated Optimizer / report_bench_2026-09-01T234338.md
Beats tuned baseline
Стенд-проверка (stage-2) · промт оператора:
(универсальный)
Ответ агента:
{ "worked": true, "confidence": 8, "verdict": "Built the semi-passive energy-gated optimizer as a local gated-Adam training loop on the matched dynamics track, using the shared rnn_small architecture and paired datasets. The idea achieved lower test MSE than the swept Adam baseline: 6.6809e-05 versus 8.2001e-05, delta=-1.5192e-05, with permutation p=0.04715, so this benchmark meets the harness criterion for a significant win. Trained-model behavior also showed nonzero active gating and positive damping energy terms, confirming the qualitative mechanism, although the original continuous-time B/(2c) plateau prediction was not directly tested as a quantitative bound.", "metrics": { "baseline": "Adam sweep best lr=0.01, weight_decay=0.0; 8-seed mean dynamics test MSE 8.200101638067281e-05; per-seed [2.028812450589612e-04,1.6224317369051278e-04,4.15411705034785e-05,4.08045707445126e-05,5.192989920033142e-05,2.7118352591060102e-05,4.964483741787262e-05,7.984488183865324e-05]", "idea": "Gated Adam best cfg lr=0.01, c=10.0, E_star=0.01, tau=0.005; 8-seed mean dynamics test MSE 6.680865521957458e-05; per-seed [1.8218360492028296e-04,9.848197805695236e-05,3.606851532822475e-05,3.849564745905809e-05,5.274467548588291e-05,2.1851565179531462e-05,5.2063427574466914e-05,5.257982775219716e-05]", "delta_mean": -1.5192361161098233e-05, "idea_wins": 6, "p_value": 0.04715 }, "bench_report": { "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.01, "weight_decay": 0.0 }, "sweep_grid_lrs": [ 0.001, 0.003, 0.01 ], "swept_weight_decay": [ 0.0, 0.0001, 0.001 ], "full_mean": 8.200101638067281e-05, "full_std": 5.487234136562193e-05, "full_per_seed": [ 0.0002028812450589612, 0.00016224317369051278, 4.15411705034785e-05, 4.08045707445126e-05, 5.192989920033142e-05, 2.7118352591060102e-05, 4.964483741787262e-05, 7.984488183865324e-05 ], "idea_grid": [ { "lr": 0.001, "c": 2.0 }, { "lr": 0.001, "c": 5.0 }, { "lr": 0.001, "c": 10.0 }, { "lr": 0.003, "c": 2.0 }, { "lr": 0.003, "c": 5.0 }, { "lr": 0.003, "c": 10.0 }, { "lr": 0.01, "c": 2.0 }, { "lr": 0.01, "c": 5.0 }, { "lr": 0.01, "c": 10.0 } ], "idea_best_cfg": { "lr": 0.01, "c": 10.0 } }, "idea": { "best_cfg": { "lr": 0.01, "c": 10.0 }, "mean": 6.680865521957458e-05, "std": 5.376186017983789e-05, "per_seed": [ 0.00018218360492028296, 9.848197805695236e-05, 3.606851532822475e-05, 3.849564745905809e-05, 5.274467548588291e-05, 2.1851565179531462e-05, 5.2063427574466914e-05, 5.257982775219716e-05 ], "n": 8 }, "comparison": { "delta_mean": -1.5192361161098233e-05, "idea_wins": 6, "n_pairs": 8, "per_seed_diffs": [ -2.0697640138678253e-05, -6.376119563356042e-05, -5.472655175253749e-06, -2.3089232854545116e-06, 8.147762855514884e-07, -5.2667874115286395e-06, 2.4185901565942913e-06, -2.7265054086456075e-05 ], "p_value": 0.04715, "mde": 1.8590723573088402e-05, "mde_rel_pct": 22.671333104927385, "verdict": "idea better (significant)", "system_worked": true }, "mechanism_signature": { "E_star": 0.01, "tau": 0.005, "c": 10.0, "transition_width_10_to_90": 0.0219722457735, "mean_active_fraction": 0.17375, "mean_gate": 0.280572488, "mean_abs_grad_work": 0.009492, "mean_damping_energy_term": 0.41718, "mean_bound_proxy_B_over_2c": 0.001445, "confirmed": true } }, "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_experiment.py", "files": [ "bench_experiment.py", "bench_report.json", "run_output.txt" ], "limitations": "The baseline is Adam rather than momentum SGD because bench.train_model defines Adam as the canonical standard path. The experiment used 1200 training and 400 test samples, 15 epochs, and the built-in dynamics prediction task; it did not test full-scale datasets, momentum-SGD integration, the proposed online exponential moving maximum for B, or a direct quantitative continuous-time ultimate-bound calibration. The idea implementation uses bias-corrected Adam first-moment/preconditioned velocity with multiplicative gated damping, so it tests the implementation-plan Adam variant rather than raw M=I momentum.", "system_verdict": "worked", "practical_verdict": "helps", "mechanism_ok": 1, "system_judged": true }