r-Deformed Power Divergence Loss / results.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
 1{
 2  "seed": 346,
 3  "alpha": 0.5,
 4  "r_candidates": [
 5    0,
 6    0.5,
 7    1,
 8    1.5
 9  ],
10  "parameterization": "commuting diagonal specialization; label smoothing=0.1; Adam; 35 epochs; batch=96",
11  "math_check": {
12    "r_values": [
13      0.16517196595668793,
14      0.17075106501579285,
15      0.1765032261610031,
16      0.17662368714809418,
17      0.1767442524433136,
18      0.18280813097953796
19    ],
20    "max_near_one_difference": 0.00012056529521942139,
21    "renyi_direct_difference": 0.0,
22    "onehot_loss_at_p": {
23      "0": 1.7999999523162842,
24      "0.5": 2.73508882522583,
25      "1": 4.605170249938965,
26      "1.5": 8.649110794067383
27    }
28  },
29  "results": {
30    "ce": {
31      "acc": 0.8600000143051147,
32      "ece": 0.07694082458754009,
33      "val_ce": 0.6132366061210632,
34      "last_loss": 0.5947586643068414,
35      "grad_mean": 0.1309836793214755,
36      "grad_var": 0.006057037108933663,
37      "grad_max": 1.3817980289459229
38    },
39    "renyi": {
40      "acc": 0.8650000095367432,
41      "ece": 0.06941253723499027,
42      "val_ce": 0.6062193512916565,
43      "last_loss": 0.12380720400496532,
44      "grad_mean": 0.08105734429581272,
45      "grad_var": 0.0017172855145267901,
46      "grad_max": 0.6620520949363708
47    },
48    "r0": {
49      "acc": 0.8650000095367432,
50      "ece": 0.06264462541998611,
51      "val_ce": 0.6056239604949951,
52      "last_loss": 0.10561453315772508,
53      "grad_mean": 0.06446286760746761,
54      "grad_var": 0.0011362680416997317,
55      "grad_max": 0.5619319677352905
56    },
57    "r05": {
58      "acc": 0.8600000143051147,
59      "ece": 0.08247189852049078,
60      "val_ce": 0.6032678484916687,
61      "last_loss": 0.11445811900653337,
62      "grad_mean": 0.06948142177740434,
63      "grad_var": 0.001029135555783028,
64      "grad_max": 0.5473694801330566
65    }
66  }
67}