Global-statistics context shortcut / bench_report.json
Mechanism confirmed, baseline not beaten
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": 0.003,
10 "weight_decay": 0.0
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "weight_decay": 0.0
17 },
18 "mean": 0.73499995470047
19 },
20 {
21 "cfg": {
22 "lr": 0.003,
23 "weight_decay": 0.0
24 },
25 "mean": 0.7299999892711639
26 },
27 {
28 "cfg": {
29 "lr": 0.001,
30 "weight_decay": 0.0001
31 },
32 "mean": 0.739999994635582
33 }
34 ],
35 "full": {
36 "mean": 0.7543749883770943,
37 "std": 0.05033869289963452,
38 "per_seed": [
39 0.6850000023841858,
40 0.7199999690055847,
41 0.824999988079071,
42 0.6899999976158142,
43 0.7649999856948853,
44 0.8149999976158142,
45 0.7899999618530273,
46 0.7450000047683716
47 ],
48 "n": 8
49 }
50 },
51 "idea": {
52 "mean": 0.7443749830126762,
53 "std": 0.05204790338215126,
54 "per_seed": [
55 0.7450000047683716,
56 0.7549999952316284,
57 0.8499999642372131,
58 0.7299999594688416,
59 0.6499999761581421,
60 0.7649999856948853,
61 0.7149999737739563,
62 0.7450000047683716
63 ],
64 "n": 8
65 },
66 "comparison": {
67 "delta_mean": -0.01000000536441803,
68 "idea_wins": 3,
69 "n_pairs": 8,
70 "per_seed_diffs": [
71 0.06000000238418579,
72 0.0350000262260437,
73 0.02499997615814209,
74 0.039999961853027344,
75 -0.11500000953674316,
76 -0.050000011920928955,
77 -0.07499998807907104,
78 0.0
79 ],
80 "p_value": 0.67055,
81 "mde": 0.05250391180418798,
82 "mde_rel_pct": 6.959922135958001,
83 "verdict": "no significant win",
84 "system_worked": false
85 },
86 "mechanism_signature": {
87 "signature": {
88 "location": "trained global-normalization layer on bench vision model",
89 "predicted_offdiagonal": -0.014527199789881706,
90 "observed_offdiagonal": -0.01452720072120428,
91 "absolute_error": 9.313225746154785e-10,
92 "n": 1024,
93 "confirmed": true
94 },
95 "idea_sweep": [
96 {
97 "cfg": {
98 "lr": 0.001,
99 "weight_decay": 0.0
100 },
101 "result": {
102 "mean": 0.7443749830126762,
103 "std": 0.05204790338215126,
104 "per_seed": [
105 0.7450000047683716,
106 0.7549999952316284,
107 0.8499999642372131,
108 0.7299999594688416,
109 0.6499999761581421,
110 0.7649999856948853,
111 0.7149999737739563,
112 0.7450000047683716
113 ],
114 "n": 8
115 }
116 },
117 {
118 "cfg": {
119 "lr": 0.003,
120 "weight_decay": 0.0
121 },
122 "result": {
123 "mean": 0.784374974668026,
124 "std": 0.0535862355865634,
125 "per_seed": [
126 0.7799999713897705,
127 0.7649999856948853,
128 0.8799999952316284,
129 0.7199999690055847,
130 0.7599999904632568,
131 0.85999995470047,
132 0.7799999713897705,
133 0.7299999594688416
134 ],
135 "n": 8
136 }
137 },
138 {
139 "cfg": {
140 "lr": 0.001,
141 "weight_decay": 0.0001
142 },
143 "result": {
144 "mean": 0.7474999725818634,
145 "std": 0.056623754264485855,
146 "per_seed": [
147 0.7199999690055847,
148 0.7649999856948853,
149 0.85999995470047,
150 0.7249999642372131,
151 0.6499999761581421,
152 0.7849999666213989,
153 0.7249999642372131,
154 0.75
155 ],
156 "n": 8
157 }
158 }
159 ],
160 "matched_architecture": true,
161 "normalization": "per-example channelwise spatial global statistics"
162 },
163 "custom_track": null
164}