Invariant nonstandard residual blocks / results.json

✓✓ Beats tuned baseline

Raw ⬇ ZIP
  1{
  2  "device": "cpu",
  3  "math_check": {
  4    "h": [
  5      0.5,
  6      1.0,
  7      2.0,
  8      4.0,
  9      8.0,
 10      32.0
 11    ],
 12    "q": [
 13      0.4,
 14      0.6666666666666666,
 15      1.0,
 16      1.3333333333333333,
 17      1.6,
 18      1.8823529411764706
 19    ],
 20    "euler_amplification": [
 21      0.5,
 22      0.0,
 23      1.0,
 24      3.0,
 25      7.0,
 26      31.0
 27    ],
 28    "nonstandard_amplification": [
 29      0.6,
 30      0.33333333333333337,
 31      0.0,
 32      0.33333333333333326,
 33      0.6000000000000001,
 34      0.8823529411764706
 35    ],
 36    "q_bound_1_over_alpha": 2.0,
 37    "nonstandard_nonexpansive_all_tested": true,
 38    "euler_nonexpansive_all_tested": false
 39  },
 40  "runs": {
 41    "baseline_h0.5": {
 42      "loss": 0.170854389667511,
 43      "accuracy": 0.9555555582046509,
 44      "max_activation": 35.38071823120117,
 45      "max_update_proxy": 3.1730897426605225,
 46      "max_grad": 2.3008549213409424,
 47      "failed_solves": 0,
 48      "nan": false,
 49      "seconds": 5.916985750198364,
 50      "parameters": 67722
 51    },
 52    "idea_h0.5": {
 53      "loss": 0.17301425337791443,
 54      "accuracy": 0.9644444584846497,
 55      "max_activation": 34.1579475402832,
 56      "max_update_proxy": 2.5038013458251953,
 57      "max_grad": 1.8154431581497192,
 58      "failed_solves": 0,
 59      "nan": false,
 60      "seconds": 18.450348138809204,
 61      "parameters": 67722
 62    },
 63    "baseline_h1.0": {
 64      "loss": 0.19068703055381775,
 65      "accuracy": 0.9666666388511658,
 66      "max_activation": 42.3558349609375,
 67      "max_update_proxy": 4.954819679260254,
 68      "max_grad": 4.357857704162598,
 69      "failed_solves": 0,
 70      "nan": false,
 71      "seconds": 5.960226774215698,
 72      "parameters": 67722
 73    },
 74    "idea_h1.0": {
 75      "loss": 0.20936831831932068,
 76      "accuracy": 0.95333331823349,
 77      "max_activation": 39.89057540893555,
 78      "max_update_proxy": 3.8923239707946777,
 79      "max_grad": 2.588390827178955,
 80      "failed_solves": 0,
 81      "nan": false,
 82      "seconds": 17.790390253067017,
 83      "parameters": 67722
 84    },
 85    "baseline_h2.0": {
 86      "loss": 0.17421558499336243,
 87      "accuracy": 0.9622222185134888,
 88      "max_activation": 55.33295822143555,
 89      "max_update_proxy": 9.214510917663574,
 90      "max_grad": 14.476286888122559,
 91      "failed_solves": 0,
 92      "nan": false,
 93      "seconds": 5.78072714805603,
 94      "parameters": 67722
 95    },
 96    "idea_h2.0": {
 97      "loss": 0.28686821460723877,
 98      "accuracy": 0.9466666579246521,
 99      "max_activation": 43.98122787475586,
100      "max_update_proxy": 4.953287601470947,
101      "max_grad": 3.8520748615264893,
102      "failed_solves": 0,
103      "nan": false,
104      "seconds": 17.982959747314453,
105      "parameters": 67722
106    },
107    "baseline_h4.0": {
108      "loss": 0.25328102707862854,
109      "accuracy": 0.9511111378669739,
110      "max_activation": 100.96012115478516,
111      "max_update_proxy": 18.89162254333496,
112      "max_grad": 43.556922912597656,
113      "failed_solves": 0,
114      "nan": false,
115      "seconds": 6.288129806518555,
116      "parameters": 67722
117    },
118    "idea_h4.0": {
119      "loss": 0.1892751157283783,
120      "accuracy": 0.9599999785423279,
121      "max_activation": 44.65541076660156,
122      "max_update_proxy": 6.599419593811035,
123      "max_grad": 6.101863384246826,
124      "failed_solves": 0,
125      "nan": false,
126      "seconds": 17.875237226486206,
127      "parameters": 67722
128    }
129  }
130}