Frequency-Response Regularization for Neural Dynamics / report_bench_2026-09-04T161741.md

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP

Стенд-проверка (stage-2) · промт оператора:

(универсальный)

Ответ агента:

{ "worked": false, "confidence": 9, "verdict": "Built frequency-response regularization for the built-in controlled-pendulum dynamics track using the shared rnn_small architecture and a custom training loop, with exact local GRU Jacobians and discrete resolvent soft-max penalty. Across 8 paired seeds, the idea had slightly worse MSE (0.0014455 vs 0.0014107; delta +0.0000348), with p=0.98335, so there is no significant task-metric win. The mechanism signature was confirmed: mean trained-model resolvent peak decreased from 1.1065 to 1.0917 (1.34%), but this did not translate into better benchmark accuracy.", "metrics": { "baseline": "test MSE mean 0.0014106978, std 0.0003818410; best config lr=0.01, weight_decay=0.0001; 8 per-seed results", "idea": "test MSE mean 0.0014454917, std 0.0009746414; best config lr=0.01, lambda_fr=0.0005; 8 per-seed results", "paired_delta": "idea - baseline = +0.0000347938 MSE", "permutation_p": "0.98335", "mechanism_signature": "trained-model local resolvent peak baseline 1.1064540, idea 1.0916701, reduction 1.336%, confirmed=true" }, "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_frequency_dynamics.py", "files": [ "bench_frequency_dynamics.py", "bench_report.json" ], "limitations": "Only the built-in dynamics track was tested, with 400 training and 200 test examples, 12 epochs, a 12-frequency regularization grid, and 8 paired seeds. The Jacobian signature was evaluated at zero hidden state and zero input rather than across a distribution of rollout states; no continuous-time neural ODE, spectral-radius control baseline, or longer-horizon robustness metric was tested. The regularizer implementation uses exact GRU Jacobians and therefore is costly for larger hidden dimensions.", "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": 0.0001 }, "sweep": [ { "cfg": { "lr": 0.001, "weight_decay": 0.0 }, "mean": 0.004298779065720737 }, { "cfg": { "lr": 0.001, "weight_decay": 0.0001 }, "mean": 0.0042995543335564435 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0 }, "mean": 0.0034615940821822733 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0001 }, "mean": 0.003432174737099558 }, { "cfg": { "lr": 0.01, "weight_decay": 0.0 }, "mean": 0.0014865802077110857 }, { "cfg": { "lr": 0.01, "weight_decay": 0.0001 }, "mean": 0.0014641905727330595 } ], "full": { "mean": 0.0014106978487689048, "std": 0.00038184104551606325, "per_seed": [ 0.0009313806658610702, 0.0018688973505049944, 0.0010692982468754053, 0.0019871860276907682, 0.0009491611272096634, 0.0014028067234903574, 0.0016651722835376859, 0.0014116803649812937 ], "n": 8 } }, "idea": { "mean": 0.0014454916745307855, "std": 0.0009746414116167044, "per_seed": [ 0.003880624659359455, 0.0016319446731358767, 0.0007165370043367147, 0.001440238207578659, 0.0013295277021825314, 0.0009956876747310162, 0.000869154348038137, 0.0007002191268838942 ], "n": 8 }, "comparison": { "delta_mean": 3.4793825761880726e-05, "idea_wins": 6, "n_pairs": 8, "per_seed_diffs": [ 0.002949243993498385, -0.00023695267736911774, -0.00035276124253869057, -0.0005469478201121092, 0.00038036657497286797, -0.00040711904875934124, -0.0007960179354995489, -0.0007114612380973995 ], "p_value": 0.98335, "mde": 0.0010298298032408945, "mde_rel_pct": 73.00144422419102, "verdict": "no significant win", "system_worked": false }, "idea_sweep": [ { "cfg": { "lr": 0.01, "lambda_fr": 0.0 }, "mean": 0.001941849637660198 }, { "cfg": { "lr": 0.01, "lambda_fr": 0.0001 }, "mean": 0.0019368745561223477 }, { "cfg": { "lr": 0.01, "lambda_fr": 0.0005 }, "mean": 0.0019173361361026764 } ], "protocol_note": "baseline sweep used all shared learning rates and weight-decay values", "mechanism_signature": { "quantity": "trained-model local discrete resolvent peak", "predicted_effect": "frequency penalty lowers peak", "baseline_peak_mean": 1.1064540445804596, "idea_peak_mean": 1.0916701406240463, "reduction_fraction": 0.013361516484870337, "confirmed": true } }, "system_verdict": "partial", "practical_verdict": "inconclusive", "mechanism_ok": 1, "system_judged": true }