Gram-Whitened Directional Pooling / bench_report.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "vision",
  4  "model": "cnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.001,
 10      "epochs": 8,
 11      "epsilon": 0.25,
 12      "lam": 0.01
 13    },
 14    "sweep": [
 15      {
 16        "cfg": {
 17          "lr": 0.001,
 18          "epochs": 8,
 19          "epsilon": 0.25,
 20          "lam": 0.01
 21        },
 22        "mean": 0.7974999696016312
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.003,
 27          "epochs": 8,
 28          "epsilon": 0.25,
 29          "lam": 0.01
 30        },
 31        "mean": 0.8012499809265137
 32      },
 33      {
 34        "cfg": {
 35          "lr": 0.006,
 36          "epochs": 8,
 37          "epsilon": 0.25,
 38          "lam": 0.01
 39        },
 40        "mean": 0.8087499737739563
 41      }
 42    ],
 43    "full": {
 44      "mean": 0.794999971985817,
 45      "std": 0.020916503869207948,
 46      "per_seed": [
 47        0.7799999713897705,
 48        0.7999999523162842,
 49        0.824999988079071,
 50        0.7849999666213989,
 51        0.8299999833106995,
 52        0.7649999856948853,
 53        0.7899999618530273,
 54        0.7849999666213989
 55      ],
 56      "n": 8
 57    }
 58  },
 59  "idea": {
 60    "mean": 0.7981249764561653,
 61    "std": 0.037411342897785256,
 62    "per_seed": [
 63      0.7799999713897705,
 64      0.8549999594688416,
 65      0.8549999594688416,
 66      0.7549999952316284,
 67      0.8100000023841858,
 68      0.7549999952316284,
 69      0.7749999761581421,
 70      0.7999999523162842
 71    ],
 72    "n": 8
 73  },
 74  "comparison": {
 75    "delta_mean": 0.003125004470348358,
 76    "idea_wins": 4,
 77    "n_pairs": 8,
 78    "per_seed_diffs": [
 79      0.0,
 80      0.05500000715255737,
 81      0.029999971389770508,
 82      -0.029999971389770508,
 83      -0.019999980926513672,
 84      -0.009999990463256836,
 85      -0.014999985694885254,
 86      0.014999985694885254
 87    ],
 88    "p_value": 0.8129,
 89    "mde": 0.02385367502560343,
 90    "mde_rel_pct": 3.0004623731016924,
 91    "verdict": "no measurable effect",
 92    "system_worked": false
 93  },
 94  "mechanism_signature": {
 95    "predicted_pou_error~0": 1.1920928955078125e-07,
 96    "observed_pooled_energy": 0.08952471613883972,
 97    "observed_ridge_projected_energy": 0.07936406880617142,
 98    "prediction_projected_le_raw": true,
 99    "confirmed": true,
100    "idea_sweep": [
101      {
102        "cfg": {
103          "lr": 0.001,
104          "epochs": 8,
105          "epsilon": 0.25,
106          "lam": 0.01
107        },
108        "res": {
109          "mean": 0.844374991953373,
110          "std": 0.02256068428063704,
111          "per_seed": [
112            0.8100000023841858,
113            0.8650000095367432,
114            0.875,
115            0.8650000095367432,
116            0.8149999976158142,
117            0.8299999833106995,
118            0.8499999642372131,
119            0.8449999690055847
120          ],
121          "n": 8
122        }
123      },
124      {
125        "cfg": {
126          "lr": 0.003,
127          "epochs": 8,
128          "epsilon": 0.35,
129          "lam": 0.01
130        },
131        "res": {
132          "mean": 0.7981249764561653,
133          "std": 0.037411342897785256,
134          "per_seed": [
135            0.7799999713897705,
136            0.8549999594688416,
137            0.8549999594688416,
138            0.7549999952316284,
139            0.8100000023841858,
140            0.7549999952316284,
141            0.7749999761581421,
142            0.7999999523162842
143          ],
144          "n": 8
145        }
146      },
147      {
148        "cfg": {
149          "lr": 0.006,
150          "epochs": 8,
151          "epsilon": 0.25,
152          "lam": 0.01
153        },
154        "res": {
155          "mean": 0.8024999797344208,
156          "std": 0.04782781963257417,
157          "per_seed": [
158            0.7799999713897705,
159            0.8999999761581421,
160            0.8100000023841858,
161            0.7149999737739563,
162            0.8100000023841858,
163            0.8199999928474426,
164            0.7949999570846558,
165            0.7899999618530273
166          ],
167          "n": 8
168        }
169      }
170    ],
171    "structural_match": "vision CNN spatial directional pooling; same task and trained systems"
172  }
173}