Proximal-Mismatch Fine-Tuning / results.json
Failed on benchmark
1{
2 "seed": 938,
3 "math_check": {
4 "coefficient_max_abs_error": 3.3306690738754696e-16,
5 "coefficient_rows": [
6 {
7 "gamma": 0.1,
8 "lambda": 0.0,
9 "pred": 1.0,
10 "fit": 1.0,
11 "abs_err": 0.0
12 },
13 {
14 "gamma": 0.1,
15 "lambda": 0.5,
16 "pred": 0.9523809523809523,
17 "fit": 0.952380952380952,
18 "abs_err": 3.3306690738754696e-16
19 },
20 {
21 "gamma": 0.1,
22 "lambda": 2.0,
23 "pred": 0.8333333333333334,
24 "fit": 0.8333333333333336,
25 "abs_err": 2.220446049250313e-16
26 },
27 {
28 "gamma": 0.1,
29 "lambda": 8.0,
30 "pred": 0.5555555555555556,
31 "fit": 0.5555555555555555,
32 "abs_err": 1.1102230246251565e-16
33 },
34 {
35 "gamma": 0.3,
36 "lambda": 0.0,
37 "pred": 1.0,
38 "fit": 1.0,
39 "abs_err": 0.0
40 },
41 {
42 "gamma": 0.3,
43 "lambda": 0.5,
44 "pred": 0.8695652173913044,
45 "fit": 0.8695652173913043,
46 "abs_err": 1.1102230246251565e-16
47 },
48 {
49 "gamma": 0.3,
50 "lambda": 2.0,
51 "pred": 0.625,
52 "fit": 0.6249999999999999,
53 "abs_err": 1.1102230246251565e-16
54 },
55 {
56 "gamma": 0.3,
57 "lambda": 8.0,
58 "pred": 0.29411764705882354,
59 "fit": 0.29411764705882354,
60 "abs_err": 0.0
61 },
62 {
63 "gamma": 0.7,
64 "lambda": 0.0,
65 "pred": 1.0,
66 "fit": 1.0,
67 "abs_err": 0.0
68 },
69 {
70 "gamma": 0.7,
71 "lambda": 0.5,
72 "pred": 0.7407407407407407,
73 "fit": 0.7407407407407408,
74 "abs_err": 1.1102230246251565e-16
75 },
76 {
77 "gamma": 0.7,
78 "lambda": 2.0,
79 "pred": 0.4166666666666667,
80 "fit": 0.4166666666666667,
81 "abs_err": 0.0
82 },
83 {
84 "gamma": 0.7,
85 "lambda": 8.0,
86 "pred": 0.15151515151515152,
87 "fit": 0.15151515151515155,
88 "abs_err": 2.7755575615628914e-17
89 },
90 {
91 "gamma": 1.0,
92 "lambda": 0.0,
93 "pred": 1.0,
94 "fit": 1.0,
95 "abs_err": 0.0
96 },
97 {
98 "gamma": 1.0,
99 "lambda": 0.5,
100 "pred": 0.6666666666666666,
101 "fit": 0.6666666666666666,
102 "abs_err": 0.0
103 },
104 {
105 "gamma": 1.0,
106 "lambda": 2.0,
107 "pred": 0.3333333333333333,
108 "fit": 0.3333333333333333,
109 "abs_err": 0.0
110 },
111 {
112 "gamma": 1.0,
113 "lambda": 8.0,
114 "pred": 0.1111111111111111,
115 "fit": 0.11111111111111112,
116 "abs_err": 1.3877787807814457e-17
117 }
118 ],
119 "mismatch_loglog_slope": 2.000000000000007,
120 "mismatch_quadratic_prediction": 2.0,
121 "lambda_zero_mismatch": 0.0,
122 "stability_predicted_boundary": 2.0,
123 "largest_stable_grid_a": 1.99,
124 "smallest_unstable_grid_a": 2.01,
125 "stability_rows": [
126 {
127 "a": 0.5,
128 "predicted_q": 0.25,
129 "observed_final_abs": 3.552713678800501e-15,
130 "stable_observed": true
131 },
132 {
133 "a": 1.0,
134 "predicted_q": 0.5,
135 "observed_final_abs": 5.960464477539063e-08,
136 "stable_observed": true
137 },
138 {
139 "a": 1.8,
140 "predicted_q": 0.9,
141 "observed_final_abs": 0.07976644307687256,
142 "stable_observed": true
143 },
144 {
145 "a": 1.99,
146 "predicted_q": 0.995,
147 "observed_final_abs": 0.8866535105013078,
148 "stable_observed": true
149 },
150 {
151 "a": 2.01,
152 "predicted_q": 1.005,
153 "observed_final_abs": 1.1271597762053889,
154 "stable_observed": false
155 },
156 {
157 "a": 2.2,
158 "predicted_q": 1.1,
159 "observed_final_abs": 9.84973267580763,
160 "stable_observed": false
161 },
162 {
163 "a": 3.0,
164 "predicted_q": 1.5,
165 "observed_final_abs": 16834.112196028233,
166 "stable_observed": false
167 }
168 ]
169 },
170 "adaptation": {
171 "a_star": 0.45454545454545453,
172 "clean_mse_coefficient": 0.6860440633405669,
173 "proximal_coefficient": 0.4545454545454545,
174 "mixed_coefficient": 0.4800983002056744,
175 "results": [
176 {
177 "method": "clean_mse",
178 "a": 0.6860440633405669,
179 "final_F": 0.4410948547609531,
180 "final_stationarity": 0.5887514386655519,
181 "reconstruction_MSE_to_optimum": 0.058122213539468136,
182 "prox_mismatch_on_states": 0.012912660641092063
183 },
184 {
185 "method": "proximal_match",
186 "a": 0.4545454545454545,
187 "final_F": 0.35391153445175094,
188 "final_stationarity": 2.0066435492405165e-16,
189 "reconstruction_MSE_to_optimum": 8.116294220563262e-33,
190 "prox_mismatch_on_states": 9.805814240963052e-34
191 },
192 {
193 "method": "mixed",
194 "a": 0.4800983002056744,
195 "final_F": 0.35476819383213837,
196 "final_stationarity": 0.05836054869135342,
197 "reconstruction_MSE_to_optimum": 0.0005711062535916371,
198 "prox_mismatch_on_states": 0.00015732491659826528
199 },
200 {
201 "method": "ideal",
202 "a": 0.45454545454545453,
203 "final_F": 0.35391153445175094,
204 "final_stationarity": 9.514546268907243e-17,
205 "reconstruction_MSE_to_optimum": 2.2369317072984132e-33,
206 "prox_mismatch_on_states": 0.0
207 }
208 ]
209 },
210 "interpretation": {
211 "mechanism_confirmed": true,
212 "practical_prox_better_than_clean": true,
213 "note": "The toy teacher is exact and quadratic; this isolates the proposed proximal-mismatch mechanism, not image-denoising generalization."
214 }
215}