FMM-Accelerated Polyharmonic Neural Field Head / bench_report.json

✓✓ Beats tuned baseline

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "spatial_anchor_regression",
  4  "model": "coordinate_mlp",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.14637889340519905
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.061082218773663044
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.012440342921763659
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.012712803669273853,
 33      "std": 0.0031444912187311223,
 34      "per_seed": [
 35        0.010250885039567947,
 36        0.014543263241648674,
 37        0.013977252878248692,
 38        0.010989970527589321,
 39        0.01702292263507843,
 40        0.01695895567536354,
 41        0.009546724148094654,
 42        0.008412455208599567
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.006530163227580488,
 49    "std": 0.0013944948580770671,
 50    "per_seed": [
 51      0.008384107612073421,
 52      0.0057844375260174274,
 53      0.006018985528498888,
 54      0.007370046339929104,
 55      0.005429707467556,
 56      0.008914164267480373,
 57      0.005202051252126694,
 58      0.005137805826961994
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -0.006182640441693366,
 64    "idea_wins": 8,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -0.001866777427494526,
 68      -0.008758825715631247,
 69      -0.007958267349749804,
 70      -0.0036199241876602173,
 71      -0.01159321516752243,
 72      -0.008044791407883167,
 73      -0.00434467289596796,
 74      -0.0032746493816375732
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.002818935574904236,
 78    "mde_rel_pct": 22.17398811654307,
 79    "verdict": "idea better (significant)",
 80    "system_worked": true
 81  },
 82  "mechanism_signature": {
 83    "prediction": "trained PHS residual should be nonzero and align with held-out error correction",
 84    "observed_mean_residual_rms": 0.11425309907644987,
 85    "observed_mean_correction_corr": 0.7042187005281448,
 86    "confirmed": true,
 87    "per_seed": [
 88      {
 89        "seed": 0,
 90        "baseline_mse": 0.010250885039567947,
 91        "idea_mse": 0.005912628024816513,
 92        "idea_residual_rms": 0.09278345853090286,
 93        "residual_target_corr": 0.6705815196037292
 94      },
 95      {
 96        "seed": 1,
 97        "baseline_mse": 0.014543264172971249,
 98        "idea_mse": 0.007037218660116196,
 99        "idea_residual_rms": 0.1361566036939621,
100        "residual_target_corr": 0.7668899893760681
101      },
102      {
103        "seed": 2,
104        "baseline_mse": 0.01397725474089384,
105        "idea_mse": 0.005881397519260645,
106        "idea_residual_rms": 0.10070868581533432,
107        "residual_target_corr": 0.7634603977203369
108      },
109      {
110        "seed": 3,
111        "baseline_mse": 0.010989971458911896,
112        "idea_mse": 0.007288732565939426,
113        "idea_residual_rms": 0.11611808091402054,
114        "residual_target_corr": 0.6222603917121887
115      },
116      {
117        "seed": 4,
118        "baseline_mse": 0.01702292263507843,
119        "idea_mse": 0.006018869113177061,
120        "idea_residual_rms": 0.10683124512434006,
121        "residual_target_corr": 0.7492429614067078
122      },
123      {
124        "seed": 5,
125        "baseline_mse": 0.01695895753800869,
126        "idea_mse": 0.00841639842838049,
127        "idea_residual_rms": 0.1272411346435547,
128        "residual_target_corr": 0.7258725762367249
129      },
130      {
131        "seed": 6,
132        "baseline_mse": 0.009546725079417229,
133        "idea_mse": 0.007395600900053978,
134        "idea_residual_rms": 0.1277274787425995,
135        "residual_target_corr": 0.7541661858558655
136      },
137      {
138        "seed": 7,
139        "baseline_mse": 0.008412455208599567,
140        "idea_mse": 0.008615097030997276,
141        "idea_residual_rms": 0.10645810514688492,
142        "residual_target_corr": 0.5812755823135376
143      }
144    ]
145  },
146  "custom_track": {
147    "name": "spatial_anchor_regression",
148    "file": "custom_spatial.py",
149    "domain": "spatial neural field / interpolation"
150  },
151  "idea_selected_cfg": {
152    "lr": 0.01
153  },
154  "idea_sweep": [
155    {
156      "cfg": {
157        "lr": 0.01
158      },
159      "mean": 0.006530163227580488,
160      "std": 0.0013944948580770671,
161      "per_seed": [
162        0.008384107612073421,
163        0.0057844375260174274,
164        0.006018985528498888,
165        0.007370046339929104,
166        0.005429707467556,
167        0.008914164267480373,
168        0.005202051252126694,
169        0.005137805826961994
170      ]
171    },
172    {
173      "cfg": {
174        "lr": 0.001
175      },
176      "mean": 0.10927551612257957,
177      "std": 0.012333453085783437,
178      "per_seed": [
179        0.12216655164957047,
180        0.12041649222373962,
181        0.12407363951206207,
182        0.10313449800014496,
183        0.08412088453769684,
184        0.10495962202548981,
185        0.11025018244981766,
186        0.10508225858211517
187      ]
188    },
189    {
190      "cfg": {
191        "lr": 0.003
192      },
193      "mean": 0.022391047328710556,
194      "std": 0.008746374453946222,
195      "per_seed": [
196        0.028721362352371216,
197        0.018648821860551834,
198        0.03950253874063492,
199        0.01322929561138153,
200        0.014868687838315964,
201        0.016263097524642944,
202        0.030357947573065758,
203        0.017536627128720284
204      ]
205    }
206  ]
207}