matCercola18 commited on
Commit
929e74d
·
verified ·
1 Parent(s): 282f14b

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. results/ablations/ablation_data.txt +18 -0
  2. results/ablations/ablation_generative_pendulum.txt +14 -0
  3. results/ablations/ablation_generative_siso.txt +18 -0
  4. results/ablations/ablation_generative_siso_pooled.txt +21 -0
  5. results/ablations/ablation_generative_siso_seed200.txt +18 -0
  6. results/ablations/ablation_generative_siso_seed300.txt +18 -0
  7. results/ablations/ablation_generative_siso_seed400.txt +18 -0
  8. results/ablations/ablation_generative_siso_seed500.txt +18 -0
  9. results/ablations/ablation_size.txt +18 -0
  10. results/ablations/ablation_steps.txt +14 -0
  11. results/ablations/ablation_steps_mimo.txt +8 -0
  12. results/ablations/pendulum_data_anomaly.txt +24 -0
  13. results/furuta_v4/ablation_interp_mlpA_raw_mlpA_ema9995_mlpA_ema999_mlp192k_ema999_mlp3.txt +11 -0
  14. results/furuta_v4/ablation_mlpB256_ema999_mlpXD_ema999.txt +3 -0
  15. results/furuta_v4/ablation_table.txt +8 -0
  16. results/furuta_v4/ablation_unetSXD_ema999.txt +2 -0
  17. results/furuta_v4/budget_curves_ddim5.txt +27 -0
  18. results/furuta_v4/budget_curves_ddpm500.txt +27 -0
  19. results/furuta_v4/budget_curves_ddpm500_gpusample.txt +27 -0
  20. results/furuta_v4/budget_reg_det_s0_n200.txt +8 -0
  21. results/furuta_v4/budget_reg_det_s1_n200.txt +8 -0
  22. results/furuta_v4/budget_reg_det_s2_n200.txt +8 -0
  23. results/furuta_v4/budget_retrN_n100.txt +8 -0
  24. results/furuta_v4/budget_retrN_n200.txt +8 -0
  25. results/furuta_v4/budget_retrN_n400.txt +8 -0
  26. results/furuta_v4/budget_timing_cpu.txt +32 -0
  27. results/furuta_v4/budget_timing_cuda.txt +32 -0
  28. results/furuta_v4/budget_unetA_ema999_ddim5_n200.txt +24 -0
  29. results/furuta_v4/budget_unetA_ema999_n200.txt +24 -0
  30. results/furuta_v4/budget_unetAseed1_ema999_ddim10_n200.txt +24 -0
  31. results/furuta_v4/budget_unetAseed1_ema999_ddim5_n200.txt +24 -0
  32. results/furuta_v4/budget_unetAseed2_ema999_ddim10_n200.txt +24 -0
  33. results/furuta_v4/budget_unetAseed2_ema999_ddim5_n200.txt +24 -0
  34. results/furuta_v4/budget_unetAseed2_ema999_n200.txt +24 -0
  35. results/furuta_v4/ddim_summary.txt +6 -0
  36. results/furuta_v4/ddim_unetA_ema999_10.txt +2 -0
  37. results/furuta_v4/ddim_unetA_ema999_5.txt +2 -0
  38. results/furuta_v4/ddim_unetA_ema999_50.txt +2 -0
  39. results/furuta_v4/ddim_unetAseed1_ema999_50.txt +2 -0
  40. results/furuta_v4/ddim_unetAseed2_ema999_10.txt +2 -0
  41. results/furuta_v4/ddim_unetAseed2_ema999_20.txt +2 -0
  42. results/furuta_v4/ddim_unetAseed2_ema999_5.txt +2 -0
  43. results/furuta_v4/ddim_unetAseed2_ema999_50.txt +2 -0
  44. results/furuta_v4/deploy_sidebyside.txt +19 -0
  45. results/furuta_v4/disjointness.txt +8 -0
  46. results/furuta_v4/gate_mlp192k_ema999.txt +2 -0
  47. results/furuta_v4/gate_mlp384k_ema9995.txt +2 -0
  48. results/furuta_v4/gate_mlpA_ema999.txt +2 -0
  49. results/furuta_v4/gate_mlpA_ema9995.txt +2 -0
  50. results/furuta_v4/gate_mlpA_raw.txt +2 -0
results/ablations/ablation_data.txt ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ === DATA ABLATION (pendulum, N=40, K=8, shared cached oracle, eval noise seed=0, 3 training seeds=[42, 43, 44]) ===
2
+
3
+ - 25% data (650/2600 examples)
4
+ params=3,823,883 time_to_best=261.8s
5
+ median regret vs oracle: +0.29% +/- 0.27pp over seeds [s42=+0.61%, s43=+0.16%, s44=+0.12%]
6
+ coverage (mean over seeds): certified>=1/plant 0.992 stable>=1/plant 1.000 per-sample stable 1.000
7
+ - 50% data (1300/2600 examples)
8
+ params=3,823,883 time_to_best=355.6s
9
+ median regret vs oracle: +1.00% +/- 0.23pp over seeds [s42=+1.22%, s43=+0.76%, s44=+1.02%]
10
+ coverage (mean over seeds): certified>=1/plant 1.000 stable>=1/plant 1.000 per-sample stable 1.000
11
+ - 100% data (= size_ref, same 3 seeds, reused)
12
+ params=3,823,883 time_to_best=421.4s
13
+ median regret vs oracle: +0.92% +/- 0.22pp over seeds [s42=+1.06%, s43=+0.67%, s44=+1.03%]
14
+ coverage (mean over seeds): certified>=1/plant 1.000 stable>=1/plant 1.000 per-sample stable 1.000
15
+ - hidden=256 (DEPLOYED checkpoint -- pre-existing train_pendulum.py run, not part of the 3-seed sweep)
16
+ params=3,823,883 time_to_best=232.0s
17
+ median regret vs oracle: +1.15% +/- 0.00pp over seeds [ssingle=+1.15%]
18
+ coverage (mean over seeds): certified>=1/plant 1.000 stable>=1/plant 1.000 per-sample stable 1.000
results/ablations/ablation_generative_pendulum.txt ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ === A3 GENERATIVE BASELINE — PENDULUM (N=40, shared cached oracle, 3 draws/K, seeds=[0, 1, 2]) ===
2
+
3
+ budget: diffusion params=3,823,883 epochs=800 time_to_best=232.0s; cvae params=277,019 epochs=800 time_to_best=278.9s
4
+
5
+ -- K=1 --
6
+ diffusion : stable>=1 1.000+-0.000 cert>=1 0.992+-0.014 persample 1.000+-0.000 median_regret +1.52%+-0.06pp diversity 0.0000+-0.0000
7
+ cvae : stable>=1 1.000+-0.000 cert>=1 1.000+-0.000 persample 1.000+-0.000 median_regret +1.37%+-0.05pp diversity 0.0000+-0.0000
8
+ paired McNemar (diffusion vs cvae, pooled 120 plant x draw trials): cvae wins 0/0 discordant, p=1
9
+
10
+ -- K=8 --
11
+ diffusion : stable>=1 1.000+-0.000 cert>=1 1.000+-0.000 persample 1.000+-0.000 median_regret +1.11%+-0.09pp diversity 0.0142+-0.0007
12
+ cvae : stable>=1 1.000+-0.000 cert>=1 1.000+-0.000 persample 1.000+-0.000 median_regret +1.20%+-0.02pp diversity 0.0027+-0.0001
13
+ paired McNemar (diffusion vs cvae, pooled 120 plant x draw trials): cvae wins 0/0 discordant, p=1
14
+
results/ablations/ablation_generative_siso.txt ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ === A3 GENERATIVE BASELINE — SISO (N=60 plants, plant_seed=100, 3 draws/K, seeds=[0, 1, 2]) ===
2
+
3
+ budget: diffusion params=21,445,653 epochs=1200 time_to_best=7940.6s; cvae params=1,108,021 epochs=60 time_to_best=33.9s; regressor params=804,373 epochs=40 time=not recorded
4
+
5
+ -- K=1 --
6
+ diffusion : plant_stabilisable 0.667+-0.033 persample 0.667+-0.033 median_best_J 6.713+-3.633 diversity 0.0000+-0.0000
7
+ cvae : plant_stabilisable 0.750+-0.050 persample 0.750+-0.050 median_best_J 5.167+-2.512 diversity 0.0000+-0.0000
8
+ regressor : plant_stabilisable 0.750+-0.044 persample 0.750+-0.044 median_best_J 2.403+-0.141 diversity 0.0000+-0.0000
9
+ paired McNemar (diffusion vs cvae, pooled 180 plant x draw trials): cvae wins 18/21 discordant, p=0.00149
10
+ paired McNemar (diffusion vs regressor, pooled 180 plant x draw trials): regressor wins 19/23 discordant, p=0.0026
11
+
12
+ -- K=8 --
13
+ diffusion : plant_stabilisable 0.800+-0.050 persample 0.640+-0.024 median_best_J 1.363+-0.120 diversity 0.1844+-0.0076
14
+ cvae : plant_stabilisable 0.767+-0.033 persample 0.751+-0.045 median_best_J 0.748+-0.113 diversity 0.0034+-0.0001
15
+ regressor : plant_stabilisable 0.750+-0.044 persample 0.750+-0.044 median_best_J 2.403+-0.141 diversity 0.0000+-0.0000
16
+ paired McNemar (diffusion vs cvae, pooled 180 plant x draw trials): cvae wins 0/6 discordant, p=0.0312
17
+ paired McNemar (diffusion vs regressor, pooled 180 plant x draw trials): regressor wins 0/9 discordant, p=0.00391
18
+
results/ablations/ablation_generative_siso_pooled.txt ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ === SISO POOLED: 5 held-out plant draws (seeds 100,200,300,400,500), n=300 plants x 3 sampler draws = 900 trials/arm/K ===
2
+ plant seeds pooled: [100, 200, 300, 400, 500]
3
+
4
+ -- K=1 --
5
+ diffusion : coverage 0.628 [0.597, 0.660] (bootstrap 95% CI, n=900 pooled trials) per-draw: seed100=0.667, seed200=0.644, seed300=0.672, seed400=0.622, seed500=0.533
6
+ cvae : coverage 0.726 [0.697, 0.756] (bootstrap 95% CI, n=900 pooled trials) per-draw: seed100=0.750, seed200=0.739, seed300=0.811, seed400=0.694, seed500=0.633
7
+ regressor : coverage 0.729 [0.700, 0.759] (bootstrap 95% CI, n=900 pooled trials) per-draw: seed100=0.750, seed200=0.750, seed300=0.806, seed400=0.700, seed500=0.639
8
+ diffusion - cvae: coverage diff -0.098 [-0.121, -0.076] McNemar b(cvae-only-success)=103 c(diffusion-only-success)=15 n_disc=118 p=2.54e-17
9
+ diffusion - regressor: coverage diff -0.101 [-0.123, -0.079] McNemar b(regressor-only-success)=103 c(diffusion-only-success)=12 n_disc=115 p=3.35e-19
10
+
11
+ -- K=8 --
12
+ diffusion : coverage 0.773 [0.746, 0.801] (bootstrap 95% CI, n=900 pooled trials) per-draw: seed100=0.800, seed200=0.756, seed300=0.839, seed400=0.761, seed500=0.711
13
+ cvae : coverage 0.748 [0.719, 0.777] (bootstrap 95% CI, n=900 pooled trials) per-draw: seed100=0.767, seed200=0.756, seed300=0.817, seed400=0.739, seed500=0.661
14
+ regressor : coverage 0.729 [0.700, 0.759] (bootstrap 95% CI, n=900 pooled trials) per-draw: seed100=0.750, seed200=0.750, seed300=0.806, seed400=0.700, seed500=0.639
15
+ diffusion - cvae: coverage diff +0.026 [+0.013, +0.039] McNemar b(cvae-only-success)=5 c(diffusion-only-success)=28 n_disc=33 p=6.62e-05
16
+ diffusion - regressor: coverage diff +0.044 [+0.030, +0.060] McNemar b(regressor-only-success)=4 c(diffusion-only-success)=44 n_disc=48 p=1.51e-09
17
+
18
+ -- K=1 -> K=8 coverage GAIN per arm (paired bootstrap CI; the mechanism claim) --
19
+ diffusion : gain +0.146 [+0.122, +0.170]
20
+ cvae : gain +0.022 [+0.013, +0.032]
21
+ regressor : gain +0.000 [+0.000, +0.000]
results/ablations/ablation_generative_siso_seed200.txt ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ === A3 GENERATIVE BASELINE — SISO (N=60 plants, plant_seed=200, 3 draws/K, seeds=[0, 1, 2]) ===
2
+
3
+ budget: diffusion params=21,445,653 epochs=1200 time_to_best=7940.6s; cvae params=1,108,021 epochs=60 time_to_best=33.9s; regressor params=804,373 epochs=40 time=not recorded
4
+
5
+ -- K=1 --
6
+ diffusion : plant_stabilisable 0.644+-0.067 persample 0.644+-0.067 median_best_J 11.448+-5.757 diversity 0.0000+-0.0000
7
+ cvae : plant_stabilisable 0.739+-0.025 persample 0.739+-0.025 median_best_J 5.993+-2.409 diversity 0.0000+-0.0000
8
+ regressor : plant_stabilisable 0.750+-0.029 persample 0.750+-0.029 median_best_J 2.380+-0.277 diversity 0.0000+-0.0000
9
+ paired McNemar (diffusion vs cvae, pooled 180 plant x draw trials): cvae wins 18/19 discordant, p=7.63e-05
10
+ paired McNemar (diffusion vs regressor, pooled 180 plant x draw trials): regressor wins 19/19 discordant, p=3.81e-06
11
+
12
+ -- K=8 --
13
+ diffusion : plant_stabilisable 0.756+-0.069 persample 0.615+-0.017 median_best_J 1.796+-0.349 diversity 0.1773+-0.0099
14
+ cvae : plant_stabilisable 0.756+-0.038 persample 0.733+-0.030 median_best_J 0.770+-0.091 diversity 0.0034+-0.0001
15
+ regressor : plant_stabilisable 0.750+-0.029 persample 0.750+-0.029 median_best_J 2.380+-0.277 diversity 0.0000+-0.0000
16
+ paired McNemar (diffusion vs cvae, pooled 180 plant x draw trials): cvae wins 2/4 discordant, p=1
17
+ paired McNemar (diffusion vs regressor, pooled 180 plant x draw trials): regressor wins 2/5 discordant, p=1
18
+
results/ablations/ablation_generative_siso_seed300.txt ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ === A3 GENERATIVE BASELINE — SISO (N=60 plants, plant_seed=300, 3 draws/K, seeds=[0, 1, 2]) ===
2
+
3
+ budget: diffusion params=21,445,653 epochs=1200 time_to_best=7940.6s; cvae params=1,108,021 epochs=60 time_to_best=33.9s; regressor params=804,373 epochs=40 time=not recorded
4
+
5
+ -- K=1 --
6
+ diffusion : plant_stabilisable 0.672+-0.054 persample 0.672+-0.054 median_best_J 7.642+-3.111 diversity 0.0000+-0.0000
7
+ cvae : plant_stabilisable 0.811+-0.035 persample 0.811+-0.035 median_best_J 5.722+-3.110 diversity 0.0000+-0.0000
8
+ regressor : plant_stabilisable 0.806+-0.025 persample 0.806+-0.025 median_best_J 2.596+-0.253 diversity 0.0000+-0.0000
9
+ paired McNemar (diffusion vs cvae, pooled 180 plant x draw trials): cvae wins 26/27 discordant, p=4.17e-07
10
+ paired McNemar (diffusion vs regressor, pooled 180 plant x draw trials): regressor wins 25/26 discordant, p=8.05e-07
11
+
12
+ -- K=8 --
13
+ diffusion : plant_stabilisable 0.839+-0.025 persample 0.681+-0.016 median_best_J 1.486+-0.162 diversity 0.1781+-0.0086
14
+ cvae : plant_stabilisable 0.817+-0.033 persample 0.806+-0.023 median_best_J 0.913+-0.223 diversity 0.0034+-0.0001
15
+ regressor : plant_stabilisable 0.806+-0.025 persample 0.806+-0.025 median_best_J 2.596+-0.253 diversity 0.0000+-0.0000
16
+ paired McNemar (diffusion vs cvae, pooled 180 plant x draw trials): cvae wins 0/4 discordant, p=0.125
17
+ paired McNemar (diffusion vs regressor, pooled 180 plant x draw trials): regressor wins 0/6 discordant, p=0.0312
18
+
results/ablations/ablation_generative_siso_seed400.txt ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ === A3 GENERATIVE BASELINE — SISO (N=60 plants, plant_seed=400, 3 draws/K, seeds=[0, 1, 2]) ===
2
+
3
+ budget: diffusion params=21,445,653 epochs=1200 time_to_best=7940.6s; cvae params=1,108,021 epochs=60 time_to_best=33.9s; regressor params=804,373 epochs=40 time=not recorded
4
+
5
+ -- K=1 --
6
+ diffusion : plant_stabilisable 0.622+-0.042 persample 0.622+-0.042 median_best_J 9.790+-5.143 diversity 0.0000+-0.0000
7
+ cvae : plant_stabilisable 0.694+-0.025 persample 0.694+-0.025 median_best_J 4.571+-1.419 diversity 0.0000+-0.0000
8
+ regressor : plant_stabilisable 0.700+-0.044 persample 0.700+-0.044 median_best_J 2.706+-0.383 diversity 0.0000+-0.0000
9
+ paired McNemar (diffusion vs cvae, pooled 180 plant x draw trials): cvae wins 20/27 discordant, p=0.0192
10
+ paired McNemar (diffusion vs regressor, pooled 180 plant x draw trials): regressor wins 20/26 discordant, p=0.00936
11
+
12
+ -- K=8 --
13
+ diffusion : plant_stabilisable 0.761+-0.035 persample 0.608+-0.017 median_best_J 1.968+-0.313 diversity 0.1875+-0.0109
14
+ cvae : plant_stabilisable 0.739+-0.042 persample 0.697+-0.031 median_best_J 0.908+-0.171 diversity 0.0034+-0.0000
15
+ regressor : plant_stabilisable 0.700+-0.044 persample 0.700+-0.044 median_best_J 2.706+-0.383 diversity 0.0000+-0.0000
16
+ paired McNemar (diffusion vs cvae, pooled 180 plant x draw trials): cvae wins 1/6 discordant, p=0.219
17
+ paired McNemar (diffusion vs regressor, pooled 180 plant x draw trials): regressor wins 1/13 discordant, p=0.00342
18
+
results/ablations/ablation_generative_siso_seed500.txt ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ === A3 GENERATIVE BASELINE — SISO (N=60 plants, plant_seed=500, 3 draws/K, seeds=[0, 1, 2]) ===
2
+
3
+ budget: diffusion params=21,445,653 epochs=1200 time_to_best=7940.6s; cvae params=1,108,021 epochs=60 time_to_best=33.9s; regressor params=804,373 epochs=40 time=not recorded
4
+
5
+ -- K=1 --
6
+ diffusion : plant_stabilisable 0.533+-0.017 persample 0.533+-0.017 median_best_J 5.136+-2.005 diversity 0.0000+-0.0000
7
+ cvae : plant_stabilisable 0.633+-0.017 persample 0.633+-0.017 median_best_J 5.881+-2.407 diversity 0.0000+-0.0000
8
+ regressor : plant_stabilisable 0.639+-0.025 persample 0.639+-0.025 median_best_J 2.357+-0.177 diversity 0.0000+-0.0000
9
+ paired McNemar (diffusion vs cvae, pooled 180 plant x draw trials): cvae wins 21/24 discordant, p=0.000277
10
+ paired McNemar (diffusion vs regressor, pooled 180 plant x draw trials): regressor wins 20/21 discordant, p=2.1e-05
11
+
12
+ -- K=8 --
13
+ diffusion : plant_stabilisable 0.711+-0.048 persample 0.537+-0.028 median_best_J 1.432+-0.271 diversity 0.1897+-0.0077
14
+ cvae : plant_stabilisable 0.661+-0.019 persample 0.635+-0.018 median_best_J 0.774+-0.171 diversity 0.0034+-0.0001
15
+ regressor : plant_stabilisable 0.639+-0.025 persample 0.639+-0.025 median_best_J 2.357+-0.177 diversity 0.0000+-0.0000
16
+ paired McNemar (diffusion vs cvae, pooled 180 plant x draw trials): cvae wins 2/13 discordant, p=0.0225
17
+ paired McNemar (diffusion vs regressor, pooled 180 plant x draw trials): regressor wins 1/15 discordant, p=0.000977
18
+
results/ablations/ablation_size.txt ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ === SIZE ABLATION (pendulum, N=40, K=8, shared cached oracle, eval noise seed=0, 3 training seeds=[42, 43, 44]) ===
2
+
3
+ - hidden=64 (~1/4 width)
4
+ params=255,563 time_to_best=405.9s
5
+ median regret vs oracle: +0.68% +/- 0.16pp over seeds [s42=+0.58%, s43=+0.87%, s44=+0.60%]
6
+ coverage (mean over seeds): certified>=1/plant 0.992 stable>=1/plant 1.000 per-sample stable 0.999
7
+ - hidden=256 (fair reference, ablation trainer)
8
+ params=3,823,883 time_to_best=421.4s
9
+ median regret vs oracle: +0.92% +/- 0.22pp over seeds [s42=+1.06%, s43=+0.67%, s44=+1.03%]
10
+ coverage (mean over seeds): certified>=1/plant 1.000 stable>=1/plant 1.000 per-sample stable 1.000
11
+ - hidden=512 (~2x width)
12
+ params=15,118,859 time_to_best=441.1s
13
+ median regret vs oracle: +0.86% +/- 0.38pp over seeds [s42=+0.98%, s43=+0.43%, s44=+1.16%]
14
+ coverage (mean over seeds): certified>=1/plant 1.000 stable>=1/plant 1.000 per-sample stable 1.000
15
+ - hidden=256 (DEPLOYED checkpoint -- pre-existing train_pendulum.py run, not part of the 3-seed sweep)
16
+ params=3,823,883 time_to_best=232.0s
17
+ median regret vs oracle: +1.15% +/- 0.00pp over seeds [ssingle=+1.15%]
18
+ coverage (mean over seeds): certified>=1/plant 1.000 stable>=1/plant 1.000 per-sample stable 1.000
results/ablations/ablation_steps.txt ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ === A4 STEP-0: DDIM@500 vs DDPM@500 quality check (3 draws each, seeds=[0, 1, 2]) ===
2
+
3
+ DDPM@500 : stable 1.000+-0.000 cert 1.000+-0.000 regret +1.11%+-0.09pp gen 1.747+-0.400 ms/ctrl
4
+ DDIM@500 : stable 1.000+-0.000 cert 1.000+-0.000 regret +1.05%+-0.05pp gen 2.218+-0.743 ms/ctrl
5
+ VERDICT: DDIM@500 MATCHES DDPM@500 within noise (|stable diff|=0.000, |regret diff|=0.06pp)
6
+
7
+ === A4 STEPS SWEEP (DDIM, eta=0, N=40, K=8, 3 draws/step-count, seeds=[0, 1, 2]) ===
8
+
9
+ steps= 5 stable>=1 1.000+-0.000 cert>=1 1.000+-0.000 persample 0.995+-0.004 median_regret +0.96%+-0.14pp wall 0.0304+-0.0069 ms/ctrl
10
+ steps= 10 stable>=1 1.000+-0.000 cert>=1 1.000+-0.000 persample 0.999+-0.002 median_regret +1.06%+-0.05pp wall 0.0362+-0.0095 ms/ctrl
11
+ steps= 20 stable>=1 1.000+-0.000 cert>=1 1.000+-0.000 persample 0.999+-0.002 median_regret +1.06%+-0.05pp wall 0.0982+-0.0376 ms/ctrl
12
+ steps= 50 stable>=1 1.000+-0.000 cert>=1 1.000+-0.000 persample 0.999+-0.002 median_regret +1.08%+-0.04pp wall 0.2618+-0.0955 ms/ctrl
13
+ steps= 100 stable>=1 1.000+-0.000 cert>=1 1.000+-0.000 persample 1.000+-0.000 median_regret +1.08%+-0.05pp wall 0.4116+-0.0402 ms/ctrl
14
+ steps= 500 stable>=1 1.000+-0.000 cert>=1 1.000+-0.000 persample 0.999+-0.002 median_regret +1.05%+-0.05pp wall 1.8847+-0.3124 ms/ctrl
results/ablations/ablation_steps_mimo.txt ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ === A4 MIMO bonus: DDPM@500 anchor + DDIM steps sweep (N=60, K=6, 3 draws, seeds=[0, 1, 2]) ===
2
+
3
+ DDPM@500 anchor: per-sample-stable 0.975+-0.006 median_regret_vs_phi0 +0.5%+-0.1 gen 3.952+-1.367 ms/ctrl
4
+
5
+ steps= 5 per-sample-stable 0.975+-0.007 has_stable>=1 1.000+-0.000 within_sector 0.890+-0.018 median_regret_vs_phi0 +1.2%+-0.5 gen 0.0357+-0.0205 ms/ctrl
6
+ steps= 20 per-sample-stable 0.972+-0.007 has_stable>=1 0.994+-0.010 within_sector 0.840+-0.013 median_regret_vs_phi0 +0.6%+-0.5 gen 0.2460+-0.2445 ms/ctrl
7
+ steps= 100 per-sample-stable 0.964+-0.007 has_stable>=1 0.994+-0.010 within_sector 0.789+-0.005 median_regret_vs_phi0 +0.4%+-0.2 gen 0.4678+-0.0237 ms/ctrl
8
+ steps= 500 per-sample-stable 0.968+-0.018 has_stable>=1 0.994+-0.010 within_sector 0.778+-0.007 median_regret_vs_phi0 +0.3%+-0.2 gen 2.6111+-0.1449 ms/ctrl
results/ablations/pendulum_data_anomaly.txt ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ === PENDULUM data-size anomaly diagnostics ===
2
+ held-out: seed123 (N=40) + seed321 (N=40), pooled N=80 plants; sampler draws (0, 1, 2); K=1 = first sample of each draw; K=8 = best of 8.
3
+ regret % vs strong per-plant oracle; failures (unstable) counted as +inf in the median, never dropped.
4
+ CI = 95% bootstrap over plants (5000) of the median pooled over (training seed x draw).
5
+
6
+ TARGET native regret (random 40 training targets, seed 2026, on own plants, K=1): median +1.41 [+1.27,+1.51] p90 +3.59 max +4.09 fail 0.000
7
+
8
+ config seeds K1 median [CI] K1 p90 K1 fail K8 median [CI] sd_seed K1 sd_seed K8 d_min d_nnplant
9
+ f0.25_e800 3 +1.16 [+0.99,+1.40] 4.18 0.000 +0.28 [+0.21,+0.37] 0.25 0.22 0.015 0.063
10
+ f0.50_e800 3 +1.48 [+1.33,+1.63] 3.06 0.000 +0.90 [+0.78,+1.04] 0.27 0.22 0.005 0.044
11
+ f1.00_e800 3 +1.32 [+1.15,+1.47] 3.09 0.000 +0.91 [+0.79,+1.03] 0.23 0.22 0.004 0.035
12
+ f0.25_e3200 2 +1.43 [+1.33,+1.58] 3.14 0.000 +1.07 [+0.95,+1.18] 0.03 0.08 0.004 0.049
13
+ f1.00_e200 2 +1.43 [+1.28,+1.63] 3.42 0.000 +0.52 [+0.45,+0.61] 0.28 0.31 0.012 0.046
14
+
15
+ per-plant paired Wilcoxon (plant = mean over seed x draw of regret; same 80 plants), two-sided:
16
+ f0.25_e800 vs f1.00_e800: K1 med diff +0.05pp p=0.146 (n=80) | K8 med diff -0.49pp p=8.46e-14 (n=80)
17
+ f0.25_e3200 vs f0.25_e800: K1 med diff +0.12pp p=0.77 (n=80) | K8 med diff +0.71pp p=3.38e-14 (n=80)
18
+ f0.25_e3200 vs f1.00_e800: K1 med diff +0.12pp p=1.59e-07 (n=80) | K8 med diff +0.12pp p=5.96e-09 (n=80)
19
+ f1.00_e200 vs f1.00_e800: K1 med diff +0.15pp p=0.00533 (n=80) | K8 med diff -0.34pp p=3.5e-11 (n=80)
20
+ f1.00_e200 vs f0.25_e800: K1 med diff +0.15pp p=0.181 (n=80) | K8 med diff +0.18pp p=1.7e-06 (n=80)
21
+ f0.25_e800 vs f0.50_e800: K1 med diff -0.11pp p=0.996 (n=80) | K8 med diff -0.48pp p=1.57e-13 (n=80)
22
+
23
+ Model K=1 vs native target regret (unpaired, different plants): compare K1 median column to the TARGET line above.
24
+ d_min = distance from a generated (clipped, normalized) controller to the nearest vector in that model's own training subset; d_nnplant = distance to the target of the nearest-condition training plant.
results/furuta_v4/ablation_interp_mlpA_raw_mlpA_ema9995_mlpA_ema999_mlp192k_ema999_mlp3.txt ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model K=1 (draw4242/draw9001, pooled, paired diff to nearest+P) // K=4
2
+ interp K=1: 60.0/61.7 pooled 60.8 vs retr 54.2 | diff +6.7 [-5.0,+17.5] p=0.2979 K=4: 70.0/71.7 pooled 70.8 vs retr 82.5 | diff -11.7 [-20.0,-3.3] p=0.0129
3
+ mlpA_raw K=1: 53.9/57.2 pooled 55.6 vs retr 54.2 | diff +1.4 [-8.3,+10.8] p=0.8250 K=4: 83.9/90.6 pooled 87.2 vs retr 82.5 | diff +4.7 [-0.8,+10.6] p=0.1459
4
+ mlpA_ema9995 K=1: 53.9/56.7 pooled 55.3 vs retr 54.2 | diff +1.1 [-8.6,+10.6] p=0.8675 K=4: 83.9/90.6 pooled 87.2 vs retr 82.5 | diff +4.7 [-0.8,+10.8] p=0.1433
5
+ mlpA_ema999 K=1: 57.2/64.4 pooled 60.8 vs retr 54.2 | diff +6.7 [-3.3,+16.4] p=0.2075 K=4: 85.6/90.0 pooled 87.8 vs retr 82.5 | diff +5.3 [-0.6,+11.4] p=0.1124
6
+ mlp192k_ema999 K=1: 60.0/73.3 pooled 66.7 vs retr 54.2 | diff +12.5 [+3.1,+21.7] p=0.0146 * K=4: 85.0/90.6 pooled 87.8 vs retr 82.5 | diff +5.3 [-0.3,+11.1] p=0.0946
7
+ mlp384k_ema9995 K=1: 66.1/80.0 pooled 73.1 vs retr 54.2 | diff +18.9 [+9.2,+28.3] p=0.0005 * K=4: 87.8/91.1 pooled 89.4 vs retr 82.5 | diff +6.9 [+1.7,+12.5] p=0.0183 *
8
+ mlp384k_ema999 K=1: 66.1/80.6 pooled 73.3 vs retr 54.2 | diff +19.2 [+8.9,+28.9] p=0.0002 * K=4: 87.8/91.1 pooled 89.4 vs retr 82.5 | diff +6.9 [+1.7,+12.5] p=0.0192 *
9
+ unetS_ema999 K=1: 88.3/92.2 pooled 90.3 vs retr 54.2 | diff +36.1 [+27.5,+45.3] p=0.0000 * K=4: 95.0/95.0 pooled 95.0 vs retr 82.5 | diff +12.5 [+7.5,+18.3] p=0.0002 *
10
+ unetA_ema9995 K=1: 90.6/93.3 pooled 91.9 vs retr 54.2 | diff +37.8 [+29.2,+46.7] p=0.0000 * K=4: 92.8/93.9 pooled 93.3 vs retr 82.5 | diff +10.8 [+5.0,+16.9] p=0.0004 *
11
+ unetA_ema999 K=1: 90.6/93.3 pooled 91.9 vs retr 54.2 | diff +37.8 [+28.9,+46.7] p=0.0000 * K=4: 92.8/93.9 pooled 93.3 vs retr 82.5 | diff +10.8 [+5.3,+16.9] p=0.0005 *
results/furuta_v4/ablation_mlpB256_ema999_mlpXD_ema999.txt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ model K=1 (draw4242/draw9001, pooled, paired diff to nearest+P) // K=4
2
+ mlpB256_ema999 K=1: 74.4/75.0 pooled 74.7 vs retr 54.2 | diff +20.6 [+10.3,+30.8] p=0.0002 * K=4: 88.3/91.7 pooled 90.0 vs retr 82.5 | diff +7.5 [+2.2,+13.1] p=0.0106 *
3
+ mlpXD_ema999 K=1: 75.6/77.8 pooled 76.7 vs retr 67.5 | diff +9.2 [+0.6,+18.1] p=0.0499 * K=4: 89.4/93.3 pooled 91.4 vs retr 90.0 | diff +1.4 [-2.2,+5.6] p=0.5889
results/furuta_v4/ablation_table.txt ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ model K=1 (draw4242/draw9001, pooled, paired diff to nearest+P) // K=4
2
+ unetAseed1_ema999 K=1: 92.8/93.9 pooled 93.3 vs retr 54.2 | diff +39.2 [+30.0,+47.8] p=0.0000 * K=4: 93.3/94.4 pooled 93.9 vs retr 82.5 | diff +11.4 [+5.6,+17.5] p=0.0006 *
3
+ unetAseed2_ema999 K=1: 90.6/93.3 pooled 91.9 vs retr 54.2 | diff +37.8 [+29.2,+46.7] p=0.0000 * K=4: 91.7/93.3 pooled 92.5 vs retr 82.5 | diff +10.0 [+4.2,+15.8] p=0.0022 *
4
+ unetA_ema999 K=1: 90.6/93.3 pooled 91.9 vs retr 54.2 | diff +37.8 [+28.6,+46.9] p=0.0000 * K=4: 92.8/93.9 pooled 93.3 vs retr 82.5 | diff +10.8 [+5.0,+17.2] p=0.0006 *
5
+ unetS_ema999 K=1: 88.3/92.2 pooled 90.3 vs retr 54.2 | diff +36.1 [+27.2,+44.7] p=0.0000 * K=4: 95.0/95.0 pooled 95.0 vs retr 82.5 | diff +12.5 [+6.7,+18.3] p=0.0001 *
6
+ unetSXD_ema999 K=1: 90.0/91.1 pooled 90.6 vs retr 67.5 | diff +23.1 [+15.0,+31.4] p=0.0000 * K=4: 95.0/95.6 pooled 95.3 vs retr 90.0 | diff +5.3 [+0.8,+10.0] p=0.0380 *
7
+ mlpB256_ema999 K=1: 74.4/75.0 pooled 74.7 vs retr 54.2 | diff +20.6 [+10.3,+30.3] p=0.0002 * K=4: 88.3/91.7 pooled 90.0 vs retr 82.5 | diff +7.5 [+1.9,+13.3] p=0.0113 *
8
+ mlpXD_ema999 K=1: 75.6/77.8 pooled 76.7 vs retr 67.5 | diff +9.2 [+0.3,+18.1] p=0.0500 K=4: 89.4/93.3 pooled 91.4 vs retr 90.0 | diff +1.4 [-2.2,+5.6] p=0.5806
results/furuta_v4/ablation_unetSXD_ema999.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ model K=1 (draw4242/draw9001, pooled, paired diff to nearest+P) // K=4
2
+ unetSXD_ema999 K=1: 90.0/91.1 pooled 90.6 vs retr 67.5 | diff +23.1 [+15.3,+31.4] p=0.0000 * K=4: 95.0/95.6 pooled 95.3 vs retr 90.0 | diff +5.3 [+0.8,+10.0] p=0.0391 *
results/furuta_v4/budget_curves_ddim5.txt ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # success (%) vs per-plant single-thread wall-clock budget; verify times as measured, costs from budget_timing_cpu.txt; refine n=200
2
+ # cost model: sample1 0.02s sample4 0.04s regfwd1 3.2ms refine(40 steps) 0.70s (flat in batch<=4)
3
+ method 0.1s 0.25s 0.5s 1s 2s 3s 5s 10s 60s best config at each budget
4
+ retrieval alone 0.0 0.0 48.3 69.2 77.5 83.3 90.0 93.3 93.3 - | - | S=16,m=1 | S=16,m=3 | S=16,m=6 | S=16,m=6 | S=16,m=7 | S=16,m=11 | S=16,m=11
5
+ retrieval + refine 0.0 0.0 48.3 83.3 87.5 93.3 97.5 97.5 97.5 - | - | S=4,m=1 | S=4,m=1,I=20 | S=4,m=2,I=40 | S=4,m=2,I=40 | S=4,m=3,I=40 | S=4,m=3,I=40 | S=4,m=3,I=40
6
+ generated (U-Net, DDIM-5) alone 0.0 0.0 80.3 81.7 82.7 92.9 93.8 94.1 94.3 - | - | S=1,m=1 | S=4,m=3 | S=8,m=5 | S=8,m=8 | S=16,m=11 | S=16,m=11 | S=16,m=13
7
+ generated (U-Net, DDIM-5) + refine 0.0 0.0 80.3 85.6 87.4 97.2 98.0 98.1 98.1 - | - | S=1,m=1 | S=1,m=1,I=20 | S=4,m=2,I=40 | S=4,m=2,I=40 | S=4,m=3,I=40 | S=4,m=4,I=40 | S=4,m=4,I=40
8
+ regression twin (deterministic) alone 0.0 0.8 76.9 76.9 76.9 85.6 85.6 85.6 85.6 - | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1
9
+ regression twin (deterministic) + refine 0.0 0.8 76.9 83.3 84.7 93.3 95.6 95.6 95.6 - | S=1,m=1 | S=1,m=1 | S=1,m=1,I=20 | S=1,m=1,I=40 | S=1,m=1,I=40 | S=1,m=1,I=40 | S=1,m=1,I=40 | S=1,m=1,I=40
10
+
11
+ # paired difference A - B in pp at each budget [95% CI] p (best config per method per budget; 120 plants)
12
+ generated (U-Net, DDIM-5) alone - retrieval alone
13
+ +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 +31.9 [+23.1,+41.0] p=0.000 +12.5 [ +5.1,+19.8] p=0.001 +5.2 [ -1.9,+12.3] p=0.165 +9.5 [ +4.3,+15.3] p=0.001 +3.8 [ +0.5, +7.9] p=0.047 +0.7 [ -2.4, +4.3] p=0.719 +0.9 [ -2.3, +4.4] p=0.622
14
+ generated (U-Net, DDIM-5) + refine - retrieval + refine
15
+ +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 +31.9 [+22.6,+41.3] p=0.000 +2.3 [ -4.4, +9.3] p=0.531 -0.1 [ -6.6, +6.8] p=1.000 +3.9 [ +0.4, +8.0] p=0.046 +0.5 [ -0.6, +2.3] p=1.000 +0.6 [ -0.4, +2.5] p=1.000 +0.6 [ -0.4, +2.5] p=1.000
16
+ generated (U-Net, DDIM-5) alone - retrieval + refine
17
+ +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 +31.9 [+23.0,+40.6] p=0.000 -1.7 [ -8.1, +5.3] p=0.658 -4.8 [-12.1, +2.1] p=0.198 -0.5 [ -4.9, +3.8] p=0.879 -3.7 [ -7.9, +0.0] p=0.086 -3.4 [ -7.5, +0.1] p=0.091 -3.2 [ -7.1, +0.4] p=0.103
18
+ generated (U-Net, DDIM-5) + refine - generated (U-Net, DDIM-5) alone
19
+ +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 +4.0 [ +1.7, +6.8] p=0.001 +4.7 [ +1.8, +8.1] p=0.002 +4.4 [ +1.6, +7.8] p=0.002 +4.2 [ +1.4, +7.4] p=0.016 +4.1 [ +1.1, +7.6] p=0.016 +3.9 [ +1.1, +7.4] p=0.015
20
+ regression twin (deterministic) alone - retrieval alone
21
+ +0.0 [ +0.0, +0.0] p=1.000 +0.8 [ +0.0, +2.5] p=1.000 +28.6 [+19.4,+38.1] p=0.000 +7.8 [ -0.0,+15.3] p=0.058 -0.6 [ -8.3, +7.2] p=0.943 +2.2 [ -3.6, +8.1] p=0.523 -4.4 [ -9.2, -0.0] p=0.085 -7.8 [-12.8, -3.1] p=0.002 -7.8 [-13.1, -3.1] p=0.002
22
+ regression twin (deterministic) alone - generated (U-Net, DDIM-5) alone
23
+ +0.0 [ +0.0, +0.0] p=1.000 +0.8 [ +0.0, +2.5] p=1.000 -3.3 [ -7.6, +0.6] p=0.131 -4.7 [ -9.4, -0.4] p=0.042 -5.7 [-10.5, -1.1] p=0.018 -7.3 [-11.7, -3.3] p=0.000 -8.2 [-12.9, -4.3] p=0.000 -8.5 [-13.1, -4.4] p=0.000 -8.7 [-13.2, -4.5] p=0.000
24
+ regression twin (deterministic) + refine - retrieval + refine
25
+ +0.0 [ +0.0, +0.0] p=1.000 +0.8 [ +0.0, +2.5] p=1.000 +28.6 [+19.2,+38.1] p=0.000 -0.0 [ -7.2, +7.2] p=1.000 -2.8 [ -9.4, +4.2] p=0.479 -0.0 [ -3.9, +3.9] p=1.000 -1.9 [ -5.3, +1.1] p=0.381 -1.9 [ -5.6, +1.1] p=0.366 -1.9 [ -5.3, +1.1] p=0.371
26
+ regression twin (deterministic) + refine - generated (U-Net, DDIM-5) + refine
27
+ +0.0 [ +0.0, +0.0] p=1.000 +0.8 [ +0.0, +2.5] p=1.000 -3.3 [ -7.6, +0.7] p=0.132 -2.3 [ -6.3, +1.6] p=0.267 -2.7 [ -6.8, +1.3] p=0.215 -3.9 [ -7.1, -1.1] p=0.014 -2.4 [ -5.3, -0.3] p=0.126 -2.6 [ -5.6, -0.4] p=0.127 -2.6 [ -5.6, -0.4] p=0.124
results/furuta_v4/budget_curves_ddpm500.txt ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # success (%) vs per-plant single-thread wall-clock budget; verify times as measured, costs from budget_timing_cpu.txt; refine n=200
2
+ # cost model: sample1 1.56s sample4 3.51s regfwd1 3.2ms refine(40 steps) 0.70s (flat in batch<=4)
3
+ method 0.1s 0.25s 0.5s 1s 2s 3s 5s 10s 60s best config at each budget
4
+ retrieval alone 0.0 0.0 48.3 69.2 77.5 83.3 90.0 93.3 93.3 - | - | S=16,m=1 | S=16,m=3 | S=16,m=6 | S=16,m=6 | S=16,m=7 | S=16,m=11 | S=16,m=11
5
+ retrieval + refine 0.0 0.0 48.3 83.3 87.5 93.3 97.5 97.5 97.5 - | - | S=4,m=1 | S=4,m=1,I=20 | S=4,m=2,I=40 | S=4,m=2,I=40 | S=4,m=3,I=40 | S=4,m=3,I=40 | S=4,m=3,I=40
6
+ generated (U-Net, DDPM-500) alone 0.0 0.0 0.0 0.0 72.1 82.5 85.8 93.0 94.2 - | - | - | - | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=4,m=4 | S=16,m=16
7
+ generated (U-Net, DDPM-500) + refine 0.0 0.0 0.0 0.0 72.1 86.5 91.1 98.0 98.4 - | - | - | - | S=1,m=1 | S=1,m=1,I=40 | S=1,m=1,I=40 | S=4,m=3,I=40 | S=4,m=4,I=40
8
+ regression twin (deterministic) alone 0.0 0.8 76.9 76.9 76.9 85.6 85.6 85.6 85.6 - | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1
9
+ regression twin (deterministic) + refine 0.0 0.8 76.9 83.3 84.7 93.3 95.6 95.6 95.6 - | S=1,m=1 | S=1,m=1 | S=1,m=1,I=20 | S=1,m=1,I=40 | S=1,m=1,I=40 | S=1,m=1,I=40 | S=1,m=1,I=40 | S=1,m=1,I=40
10
+
11
+ # paired difference A - B in pp at each budget [95% CI] p (best config per method per budget; 120 plants)
12
+ generated (U-Net, DDPM-500) alone - retrieval alone
13
+ +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 -48.3 [-57.5,-39.2] p=0.000 -69.2 [-77.5,-60.8] p=0.000 -5.4 [-12.1, +1.4] p=0.135 -0.8 [ -6.9, +5.6] p=0.821 -4.2 [ -8.8, +0.6] p=0.100 -0.4 [ -3.6, +3.1] p=0.910 +0.8 [ -2.5, +4.4] p=0.811
14
+ generated (U-Net, DDPM-500) + refine - retrieval + refine
15
+ +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 -48.3 [-57.5,-39.2] p=0.000 -83.3 [-89.2,-76.7] p=0.000 -15.4 [-22.1, -8.5] p=0.000 -6.9 [-12.5, -1.0] p=0.024 -6.4 [-10.4, -2.5] p=0.001 +0.5 [ -1.4, +2.8] p=0.749 +0.9 [ -0.9, +3.3] p=0.499
16
+ generated (U-Net, DDPM-500) alone - retrieval + refine
17
+ +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 -48.3 [-57.5,-40.0] p=0.000 -83.3 [-90.0,-75.8] p=0.000 -15.4 [-22.4, -8.6] p=0.000 -10.8 [-16.8, -5.2] p=0.000 -11.7 [-17.2, -6.6] p=0.000 -4.5 [ -8.8, -0.6] p=0.045 -3.3 [ -7.4, +0.2] p=0.140
18
+ generated (U-Net, DDPM-500) + refine - generated (U-Net, DDPM-500) alone
19
+ +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 +4.0 [ +1.5, +6.9] p=0.002 +5.3 [ +1.9, +8.9] p=0.002 +5.0 [ +1.7, +8.9] p=0.008 +4.3 [ +1.2, +8.0] p=0.014
20
+ regression twin (deterministic) alone - retrieval alone
21
+ +0.0 [ +0.0, +0.0] p=1.000 +0.8 [ +0.0, +2.5] p=1.000 +28.6 [+19.4,+38.1] p=0.000 +7.8 [ -0.0,+15.3] p=0.058 -0.6 [ -8.3, +7.2] p=0.943 +2.2 [ -3.6, +8.1] p=0.523 -4.4 [ -9.2, -0.0] p=0.085 -7.8 [-12.8, -3.1] p=0.002 -7.8 [-13.1, -3.1] p=0.002
22
+ regression twin (deterministic) alone - generated (U-Net, DDPM-500) alone
23
+ +0.0 [ +0.0, +0.0] p=1.000 +0.8 [ +0.0, +2.5] p=1.000 +76.9 [+70.0,+83.6] p=0.000 +76.9 [+70.3,+83.3] p=0.000 +4.8 [ +0.3, +9.1] p=0.041 +3.1 [ -2.4, +8.8] p=0.313 -0.3 [ -5.1, +4.4] p=0.940 -7.4 [-11.9, -3.5] p=0.000 -8.6 [-13.4, -4.4] p=0.000
24
+ regression twin (deterministic) + refine - retrieval + refine
25
+ +0.0 [ +0.0, +0.0] p=1.000 +0.8 [ +0.0, +2.5] p=1.000 +28.6 [+19.2,+38.1] p=0.000 -0.0 [ -7.2, +7.2] p=1.000 -2.8 [ -9.4, +4.2] p=0.479 -0.0 [ -3.9, +3.9] p=1.000 -1.9 [ -5.3, +1.1] p=0.381 -1.9 [ -5.6, +1.1] p=0.366 -1.9 [ -5.3, +1.1] p=0.371
26
+ regression twin (deterministic) + refine - generated (U-Net, DDPM-500) + refine
27
+ +0.0 [ +0.0, +0.0] p=1.000 +0.8 [ +0.0, +2.5] p=1.000 +76.9 [+70.3,+83.6] p=0.000 +83.3 [+77.5,+88.9] p=0.000 +12.6 [ +8.1,+17.1] p=0.000 +6.9 [ +1.9,+12.4] p=0.010 +4.4 [ +0.6, +8.4] p=0.033 -2.4 [ -5.2, -0.2] p=0.066 -2.9 [ -5.6, -0.6] p=0.062
results/furuta_v4/budget_curves_ddpm500_gpusample.txt ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # success (%) vs per-plant single-thread wall-clock budget; verify times as measured, costs from budget_timing_cpu.txt; refine n=200
2
+ # cost model: sample1 0.99s sample4 0.94s regfwd1 2.1ms refine(40 steps) 0.70s (flat in batch<=4)
3
+ method 0.1s 0.25s 0.5s 1s 2s 3s 5s 10s 60s best config at each budget
4
+ retrieval alone 0.0 0.0 48.3 69.2 77.5 83.3 90.0 93.3 93.3 - | - | S=16,m=1 | S=16,m=3 | S=16,m=6 | S=16,m=6 | S=16,m=7 | S=16,m=11 | S=16,m=11
5
+ retrieval + refine 0.0 0.0 48.3 83.3 87.5 93.3 97.5 97.5 97.5 - | - | S=4,m=1 | S=4,m=1,I=20 | S=4,m=2,I=40 | S=4,m=2,I=40 | S=4,m=3,I=40 | S=4,m=3,I=40 | S=4,m=3,I=40
6
+ generated (U-Net, DDPM-500) alone 0.0 0.0 0.0 0.0 83.1 83.2 92.9 94.0 94.2 - | - | - | - | S=4,m=2 | S=4,m=4 | S=4,m=4 | S=16,m=16 | S=16,m=16
7
+ generated (U-Net, DDPM-500) + refine 0.0 0.0 0.0 0.0 85.0 87.4 97.3 98.4 98.4 - | - | - | - | S=4,m=1,I=10 | S=1,m=1,I=40 | S=4,m=1,I=40 | S=4,m=4,I=40 | S=4,m=4,I=40
8
+ regression twin (deterministic) alone 0.0 1.1 76.9 76.9 76.9 85.6 85.6 85.6 85.6 - | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1 | S=1,m=1
9
+ regression twin (deterministic) + refine 0.0 1.1 76.9 83.3 84.7 93.3 95.6 95.6 95.6 - | S=1,m=1 | S=1,m=1 | S=1,m=1,I=20 | S=1,m=1,I=40 | S=1,m=1,I=40 | S=1,m=1,I=40 | S=1,m=1,I=40 | S=1,m=1,I=40
10
+
11
+ # paired difference A - B in pp at each budget [95% CI] p (best config per method per budget; 120 plants)
12
+ generated (U-Net, DDPM-500) alone - retrieval alone
13
+ +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 -48.3 [-57.5,-39.2] p=0.000 -69.2 [-77.5,-60.8] p=0.000 +5.6 [ -1.6,+12.8] p=0.141 -0.1 [ -6.4, +6.4] p=1.000 +2.9 [ -0.9, +7.2] p=0.195 +0.6 [ -2.4, +4.0] p=0.811 +0.8 [ -2.5, +4.4] p=0.811
14
+ generated (U-Net, DDPM-500) + refine - retrieval + refine
15
+ +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 -48.3 [-57.5,-39.2] p=0.000 -83.3 [-89.2,-76.7] p=0.000 -2.5 [ -9.4, +4.7] p=0.515 -5.9 [-11.7, -0.1] p=0.054 -0.2 [ -2.8, +2.4] p=1.000 +0.9 [ -0.9, +3.3] p=0.500 +0.9 [ -0.9, +3.3] p=0.499
16
+ generated (U-Net, DDPM-500) alone - retrieval + refine
17
+ +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 -48.3 [-57.5,-40.0] p=0.000 -83.3 [-90.0,-75.8] p=0.000 -4.4 [-11.8, +2.5] p=0.246 -10.1 [-16.2, -4.4] p=0.001 -4.6 [ -9.2, -0.5] p=0.041 -3.5 [ -7.6, +0.1] p=0.125 -3.3 [ -7.4, +0.2] p=0.140
18
+ generated (U-Net, DDPM-500) + refine - generated (U-Net, DDPM-500) alone
19
+ +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 +0.0 [ +0.0, +0.0] p=1.000 +1.9 [ +0.2, +4.3] p=0.124 +4.2 [ +1.3, +7.8] p=0.016 +4.4 [ +1.5, +7.9] p=0.016 +4.4 [ +1.2, +8.3] p=0.016 +4.3 [ +1.2, +8.0] p=0.014
20
+ regression twin (deterministic) alone - retrieval alone
21
+ +0.0 [ +0.0, +0.0] p=1.000 +1.1 [ +0.0, +3.1] p=0.502 +28.6 [+19.4,+38.1] p=0.000 +7.8 [ -0.0,+15.3] p=0.058 -0.6 [ -8.3, +7.2] p=0.943 +2.2 [ -3.6, +8.1] p=0.523 -4.4 [ -9.2, -0.0] p=0.085 -7.8 [-12.8, -3.1] p=0.002 -7.8 [-13.1, -3.1] p=0.002
22
+ regression twin (deterministic) alone - generated (U-Net, DDPM-500) alone
23
+ +0.0 [ +0.0, +0.0] p=1.000 +1.1 [ +0.0, +3.1] p=0.502 +76.9 [+70.0,+83.6] p=0.000 +76.9 [+70.3,+83.3] p=0.000 -6.1 [-10.9, -1.8] p=0.010 +2.3 [ -3.4, +8.2] p=0.470 -7.3 [-11.8, -3.3] p=0.001 -8.4 [-13.1, -4.3] p=0.000 -8.6 [-13.4, -4.4] p=0.000
24
+ regression twin (deterministic) + refine - retrieval + refine
25
+ +0.0 [ +0.0, +0.0] p=1.000 +1.1 [ +0.0, +3.1] p=0.504 +28.6 [+19.2,+38.1] p=0.000 -0.0 [ -7.2, +7.2] p=1.000 -2.8 [ -9.4, +4.2] p=0.479 -0.0 [ -3.9, +3.9] p=1.000 -1.9 [ -5.3, +1.1] p=0.381 -1.9 [ -5.6, +1.1] p=0.366 -1.9 [ -5.3, +1.1] p=0.371
26
+ regression twin (deterministic) + refine - generated (U-Net, DDPM-500) + refine
27
+ +0.0 [ +0.0, +0.0] p=1.000 +1.1 [ +0.0, +3.1] p=0.503 +76.9 [+70.3,+83.6] p=0.000 +83.3 [+77.5,+88.9] p=0.000 -0.3 [ -4.3, +3.7] p=0.932 +5.9 [ +0.9,+11.4] p=0.031 -1.8 [ -4.5, +0.4] p=0.254 -2.9 [ -5.8, -0.6] p=0.066 -2.9 [ -5.6, -0.6] p=0.062
results/furuta_v4/budget_reg_det_s0_n200.txt ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ seed4242 gen_s0 n_sub=200
2
+ [gen_s0] SDP+verify 7s plain ok 0.833; 10 failed-certified in 10 plants
3
+ [gen_s0] refine done 29s
4
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.867, 0.917, 0.933, 0.95] (K=1) (33s)
5
+ seed9001 gen_s0 n_sub=200
6
+ [gen_s0] SDP+verify 6s plain ok 0.900; 6 failed-certified in 6 plants
7
+ [gen_s0] refine done 24s
8
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.917, 0.933, 0.95, 0.967] (K=1) (27s)
results/furuta_v4/budget_reg_det_s1_n200.txt ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ seed4242 gen_s0 n_sub=200
2
+ [gen_s0] SDP+verify 7s plain ok 0.850; 9 failed-certified in 9 plants
3
+ [gen_s0] refine done 28s
4
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.9, 0.917, 0.933, 0.933] (K=1) (31s)
5
+ seed9001 gen_s0 n_sub=200
6
+ [gen_s0] SDP+verify 6s plain ok 0.867; 8 failed-certified in 8 plants
7
+ [gen_s0] refine done 26s
8
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.917, 0.95, 0.967, 0.983] (K=1) (29s)
results/furuta_v4/budget_reg_det_s2_n200.txt ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ seed4242 gen_s0 n_sub=200
2
+ [gen_s0] SDP+verify 6s plain ok 0.817; 11 failed-certified in 11 plants
3
+ [gen_s0] refine done 30s
4
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.833, 0.867, 0.917, 0.933] (K=1) (33s)
5
+ seed9001 gen_s0 n_sub=200
6
+ [gen_s0] SDP+verify 8s plain ok 0.867; 8 failed-certified in 8 plants
7
+ [gen_s0] refine done 27s
8
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.9, 0.933, 0.933, 0.967] (K=1) (31s)
results/furuta_v4/budget_retrN_n100.txt ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ seed4242 retr n_sub=100
2
+ [retr] SDP+verify 44s plain ok 0.396; 145 failed-certified in 57 plants
3
+ [retr] refine done 79s
4
+ [retr] refined ok at (5, 10, 20, 40): [0.75, 0.867, 0.883, 0.917] (K=1) (107s)
5
+ seed9001 retr n_sub=100
6
+ [retr] SDP+verify 38s plain ok 0.433; 136 failed-certified in 56 plants
7
+ [retr] refine done 57s
8
+ [retr] refined ok at (5, 10, 20, 40): [0.833, 0.9, 0.917, 0.933] (K=1) (84s)
results/furuta_v4/budget_retrN_n200.txt ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ seed4242 retr n_sub=200
2
+ [retr] SDP+verify 65s plain ok 0.322; 145 failed-certified in 57 plants
3
+ [retr] refine done 87s
4
+ [retr] refined ok at (5, 10, 20, 40): [0.767, 0.867, 0.9, 0.917] (K=1) (99s)
5
+ seed9001 retr n_sub=200
6
+ [retr] SDP+verify 56s plain ok 0.319; 136 failed-certified in 56 plants
7
+ [retr] refine done 78s
8
+ [retr] refined ok at (5, 10, 20, 40): [0.817, 0.917, 0.917, 0.967] (K=1) (89s)
results/furuta_v4/budget_retrN_n400.txt ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ seed4242 retr n_sub=400
2
+ [retr] SDP+verify 44s plain ok 0.396; 145 failed-certified in 57 plants
3
+ [retr] refine done 102s
4
+ [retr] refined ok at (5, 10, 20, 40): [0.767, 0.867, 0.9, 0.933] (K=1) (130s)
5
+ seed9001 retr n_sub=400
6
+ [retr] SDP+verify 38s plain ok 0.433; 136 failed-certified in 56 plants
7
+ [retr] refine done 78s
8
+ [retr] refined ok at (5, 10, 20, 40): [0.817, 0.917, 0.917, 0.967] (K=1) (104s)
results/furuta_v4/budget_timing_cpu.txt ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # virgo-asl load (16.4560546875, 16.083984375, 16.00244140625) device cpu
2
+ sample1 1.5558
3
+ sample4 3.5107
4
+ sample8 5.9910
5
+ sample16 10.1967
6
+ sample1_ddim5 0.0167
7
+ sample4_ddim5 0.0355
8
+ sample8_ddim5 0.0586
9
+ sample16_ddim5 0.1014
10
+ sample1_ddim10 0.0315
11
+ sample4_ddim10 0.0688
12
+ sample8_ddim10 0.1178
13
+ sample16_ddim10 0.2043
14
+ sample1_ddim20 0.0640
15
+ sample4_ddim20 0.1433
16
+ sample8_ddim20 0.2364
17
+ sample16_ddim20 0.4055
18
+ sample1_ddim50 0.1808
19
+ sample4_ddim50 0.3675
20
+ sample8_ddim50 0.6048
21
+ sample16_ddim50 1.0319
22
+ regfwd1 0.00319
23
+ regfwd8 0.01162
24
+ refine1_n100 0.366 (first call incl. compile 8.3)
25
+ refine2_n100 0.370 (first call incl. compile 0.7)
26
+ refine4_n100 0.359 (first call incl. compile 0.7)
27
+ refine1_n200 0.662 (first call incl. compile 1.0)
28
+ refine2_n200 0.689 (first call incl. compile 1.0)
29
+ refine4_n200 0.698 (first call incl. compile 1.0)
30
+ refine1_n400 1.389 (first call incl. compile 1.7)
31
+ refine2_n400 1.397 (first call incl. compile 1.8)
32
+ refine4_n400 1.454 (first call incl. compile 1.8)
results/furuta_v4/budget_timing_cuda.txt ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # virgo-asl load (10.923828125, 7.130859375, 6.99365234375) device cuda
2
+ sample1 0.9937
3
+ sample4 0.9415
4
+ sample8 1.1329
5
+ sample16 1.5277
6
+ sample1_ddim5 0.0664
7
+ sample4_ddim5 0.0675
8
+ sample8_ddim5 0.0691
9
+ sample16_ddim5 0.0718
10
+ sample1_ddim10 0.0945
11
+ sample4_ddim10 0.0808
12
+ sample8_ddim10 0.0921
13
+ sample16_ddim10 0.0287
14
+ sample1_ddim20 0.0650
15
+ sample4_ddim20 0.0989
16
+ sample8_ddim20 0.1608
17
+ sample16_ddim20 0.1395
18
+ sample1_ddim50 0.6225
19
+ sample4_ddim50 0.3997
20
+ sample8_ddim50 0.4396
21
+ sample16_ddim50 0.4600
22
+ regfwd1 0.00208
23
+ regfwd8 0.00349
24
+ refine1_n100 1.168 (first call incl. compile 7.9)
25
+ refine2_n100 1.143 (first call incl. compile 6.9)
26
+ refine4_n100 1.148 (first call incl. compile 6.9)
27
+ refine1_n200 2.030 (first call incl. compile 6.0)
28
+ refine2_n200 1.935 (first call incl. compile 7.7)
29
+ refine4_n200 1.986 (first call incl. compile 7.4)
30
+ refine1_n400 3.623 (first call incl. compile 8.0)
31
+ refine2_n400 3.598 (first call incl. compile 9.6)
32
+ refine4_n400 3.850 (first call incl. compile 9.4)
results/furuta_v4/budget_unetA_ema999_ddim5_n200.txt ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ seed4242 gen_s0 n_sub=200
2
+ [gen_s0] SDP+verify 63s plain ok 0.878; 29 failed-certified in 10 plants
3
+ [gen_s0] refine done 85s
4
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.933, 0.933, 0.95, 0.967] (K=1) (89s)
5
+ seed4242 gen_s1 n_sub=200
6
+ [gen_s1] SDP+verify 67s plain ok 0.875; 31 failed-certified in 10 plants
7
+ [gen_s1] refine done 88s
8
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.933, 0.933, 0.967, 0.983] (K=1) (93s)
9
+ seed4242 gen_s2 n_sub=200
10
+ [gen_s2] SDP+verify 61s plain ok 0.878; 33 failed-certified in 10 plants
11
+ [gen_s2] refine done 82s
12
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.95, 0.95, 0.967, 0.967] (K=1) (87s)
13
+ seed9001 gen_s0 n_sub=200
14
+ [gen_s0] SDP+verify 59s plain ok 0.933; 17 failed-certified in 5 plants
15
+ [gen_s0] refine done 75s
16
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.967, 0.967, 0.95, 0.983] (K=1) (78s)
17
+ seed9001 gen_s1 n_sub=200
18
+ [gen_s1] SDP+verify 58s plain ok 0.935; 17 failed-certified in 5 plants
19
+ [gen_s1] refine done 74s
20
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.95, 0.95, 0.933, 0.983] (K=1) (77s)
21
+ seed9001 gen_s2 n_sub=200
22
+ [gen_s2] SDP+verify 57s plain ok 0.934; 16 failed-certified in 4 plants
23
+ [gen_s2] refine done 72s
24
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.967, 0.967, 0.95, 0.983] (K=1) (76s)
results/furuta_v4/budget_unetA_ema999_n200.txt ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ seed4242 gen_s0 n_sub=200
2
+ [gen_s0] SDP+verify 71s plain ok 0.907; 23 failed-certified in 8 plants
3
+ [gen_s0] refine done 91s
4
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.933, 0.933, 0.967, 0.967] (K=1) (95s)
5
+ seed4242 gen_s1 n_sub=200
6
+ [gen_s1] SDP+verify 72s plain ok 0.902; 23 failed-certified in 7 plants
7
+ [gen_s1] refine done 90s
8
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.933, 0.933, 0.967, 0.967] (K=1) (94s)
9
+ seed4242 gen_s2 n_sub=200
10
+ [gen_s2] SDP+verify 72s plain ok 0.910; 24 failed-certified in 8 plants
11
+ [gen_s2] refine done 93s
12
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.933, 0.933, 0.967, 0.967] (K=1) (97s)
13
+ seed9001 gen_s0 n_sub=200
14
+ [gen_s0] SDP+verify 63s plain ok 0.934; 15 failed-certified in 4 plants
15
+ [gen_s0] refine done 78s
16
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.967, 0.967, 0.95, 0.983] (K=1) (81s)
17
+ seed9001 gen_s1 n_sub=200
18
+ [gen_s1] SDP+verify 64s plain ok 0.933; 16 failed-certified in 4 plants
19
+ [gen_s1] refine done 79s
20
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.967, 0.967, 0.95, 1.0] (K=1) (83s)
21
+ seed9001 gen_s2 n_sub=200
22
+ [gen_s2] SDP+verify 67s plain ok 0.933; 16 failed-certified in 4 plants
23
+ [gen_s2] refine done 82s
24
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.933, 0.967, 0.95, 1.0] (K=1) (86s)
results/furuta_v4/budget_unetAseed1_ema999_ddim10_n200.txt ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ seed4242 gen_s0 n_sub=200
2
+ [gen_s0] SDP+verify 66s plain ok 0.895; 25 failed-certified in 8 plants
3
+ [gen_s0] refine done 86s
4
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.917, 0.917, 0.967, 0.967] (K=1) (90s)
5
+ seed4242 gen_s1 n_sub=200
6
+ [gen_s1] SDP+verify 61s plain ok 0.902; 24 failed-certified in 7 plants
7
+ [gen_s1] refine done 80s
8
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.933, 0.933, 0.95, 0.967] (K=1) (84s)
9
+ seed4242 gen_s2 n_sub=200
10
+ [gen_s2] SDP+verify 66s plain ok 0.909; 22 failed-certified in 7 plants
11
+ [gen_s2] refine done 85s
12
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.933, 0.933, 0.967, 0.95] (K=1) (89s)
13
+ seed9001 gen_s0 n_sub=200
14
+ [gen_s0] SDP+verify 55s plain ok 0.940; 16 failed-certified in 5 plants
15
+ [gen_s0] refine done 71s
16
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.95, 0.967, 0.967, 0.983] (K=1) (75s)
17
+ seed9001 gen_s1 n_sub=200
18
+ [gen_s1] SDP+verify 56s plain ok 0.934; 15 failed-certified in 4 plants
19
+ [gen_s1] refine done 71s
20
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.95, 0.967, 0.967, 0.983] (K=1) (74s)
21
+ seed9001 gen_s2 n_sub=200
22
+ [gen_s2] SDP+verify 55s plain ok 0.932; 16 failed-certified in 5 plants
23
+ [gen_s2] refine done 71s
24
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.95, 0.967, 0.967, 0.983] (K=1) (75s)
results/furuta_v4/budget_unetAseed1_ema999_ddim5_n200.txt ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ seed4242 gen_s0 n_sub=200
2
+ [gen_s0] SDP+verify 62s plain ok 0.872; 28 failed-certified in 10 plants
3
+ [gen_s0] refine done 84s
4
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.917, 0.95, 0.95, 0.95] (K=1) (88s)
5
+ seed4242 gen_s1 n_sub=200
6
+ [gen_s1] SDP+verify 61s plain ok 0.878; 28 failed-certified in 9 plants
7
+ [gen_s1] refine done 82s
8
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.933, 0.933, 0.933, 0.967] (K=1) (86s)
9
+ seed4242 gen_s2 n_sub=200
10
+ [gen_s2] SDP+verify 63s plain ok 0.884; 29 failed-certified in 9 plants
11
+ [gen_s2] refine done 84s
12
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.933, 0.933, 0.95, 0.95] (K=1) (89s)
13
+ seed9001 gen_s0 n_sub=200
14
+ [gen_s0] SDP+verify 55s plain ok 0.933; 17 failed-certified in 5 plants
15
+ [gen_s0] refine done 71s
16
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.95, 0.967, 0.967, 0.983] (K=1) (75s)
17
+ seed9001 gen_s1 n_sub=200
18
+ [gen_s1] SDP+verify 58s plain ok 0.932; 17 failed-certified in 5 plants
19
+ [gen_s1] refine done 74s
20
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.933, 0.95, 0.967, 0.983] (K=1) (78s)
21
+ seed9001 gen_s2 n_sub=200
22
+ [gen_s2] SDP+verify 56s plain ok 0.929; 17 failed-certified in 5 plants
23
+ [gen_s2] refine done 72s
24
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.933, 0.967, 0.967, 0.967] (K=1) (76s)
results/furuta_v4/budget_unetAseed2_ema999_ddim10_n200.txt ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ seed4242 gen_s0 n_sub=200
2
+ [gen_s0] SDP+verify 60s plain ok 0.898; 23 failed-certified in 8 plants
3
+ [gen_s0] refine done 79s
4
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.933, 0.95, 0.967, 0.967] (K=1) (84s)
5
+ seed4242 gen_s1 n_sub=200
6
+ [gen_s1] SDP+verify 60s plain ok 0.887; 26 failed-certified in 9 plants
7
+ [gen_s1] refine done 81s
8
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.933, 0.933, 0.967, 0.967] (K=1) (85s)
9
+ seed4242 gen_s2 n_sub=200
10
+ [gen_s2] SDP+verify 61s plain ok 0.893; 27 failed-certified in 9 plants
11
+ [gen_s2] refine done 82s
12
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.95, 0.967, 0.983, 0.983] (K=1) (86s)
13
+ seed9001 gen_s0 n_sub=200
14
+ [gen_s0] SDP+verify 55s plain ok 0.935; 16 failed-certified in 4 plants
15
+ [gen_s0] refine done 70s
16
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.967, 0.967, 0.95, 0.983] (K=1) (74s)
17
+ seed9001 gen_s1 n_sub=200
18
+ [gen_s1] SDP+verify 56s plain ok 0.939; 16 failed-certified in 4 plants
19
+ [gen_s1] refine done 71s
20
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.967, 0.967, 0.983, 0.983] (K=1) (75s)
21
+ seed9001 gen_s2 n_sub=200
22
+ [gen_s2] SDP+verify 54s plain ok 0.934; 16 failed-certified in 5 plants
23
+ [gen_s2] refine done 70s
24
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.967, 0.967, 0.967, 0.983] (K=1) (73s)
results/furuta_v4/budget_unetAseed2_ema999_ddim5_n200.txt ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ seed4242 gen_s0 n_sub=200
2
+ [gen_s0] SDP+verify 62s plain ok 0.875; 30 failed-certified in 9 plants
3
+ [gen_s0] refine done 83s
4
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.9, 0.933, 0.95, 0.983] (K=1) (88s)
5
+ seed4242 gen_s1 n_sub=200
6
+ [gen_s1] SDP+verify 61s plain ok 0.878; 28 failed-certified in 10 plants
7
+ [gen_s1] refine done 82s
8
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.95, 0.95, 0.983, 0.983] (K=1) (87s)
9
+ seed4242 gen_s2 n_sub=200
10
+ [gen_s2] SDP+verify 62s plain ok 0.880; 29 failed-certified in 9 plants
11
+ [gen_s2] refine done 83s
12
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.95, 0.967, 0.983, 0.983] (K=1) (88s)
13
+ seed9001 gen_s0 n_sub=200
14
+ [gen_s0] SDP+verify 57s plain ok 0.936; 16 failed-certified in 5 plants
15
+ [gen_s0] refine done 72s
16
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.95, 0.967, 0.967, 0.983] (K=1) (76s)
17
+ seed9001 gen_s1 n_sub=200
18
+ [gen_s1] SDP+verify 56s plain ok 0.935; 15 failed-certified in 4 plants
19
+ [gen_s1] refine done 71s
20
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.967, 0.967, 0.983, 0.983] (K=1) (75s)
21
+ seed9001 gen_s2 n_sub=200
22
+ [gen_s2] SDP+verify 53s plain ok 0.931; 16 failed-certified in 5 plants
23
+ [gen_s2] refine done 69s
24
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.967, 0.967, 0.967, 0.983] (K=1) (73s)
results/furuta_v4/budget_unetAseed2_ema999_n200.txt ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ seed4242 gen_s0 n_sub=200
2
+ [gen_s0] SDP+verify 103s plain ok 0.909; 21 failed-certified in 6 plants
3
+ [gen_s0] refine done 129s
4
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.933, 0.95, 0.967, 0.967] (K=1) (135s)
5
+ seed4242 gen_s1 n_sub=200
6
+ [gen_s1] SDP+verify 107s plain ok 0.905; 23 failed-certified in 8 plants
7
+ [gen_s1] refine done 136s
8
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.933, 0.933, 0.967, 0.967] (K=1) (143s)
9
+ seed4242 gen_s2 n_sub=200
10
+ [gen_s2] SDP+verify 100s plain ok 0.914; 21 failed-certified in 6 plants
11
+ [gen_s2] refine done 124s
12
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.933, 0.95, 0.967, 0.967] (K=1) (130s)
13
+ seed9001 gen_s0 n_sub=200
14
+ [gen_s0] SDP+verify 90s plain ok 0.935; 16 failed-certified in 4 plants
15
+ [gen_s0] refine done 112s
16
+ [gen_s0] refined ok at (5, 10, 20, 40): [0.95, 0.967, 0.967, 1.0] (K=1) (118s)
17
+ seed9001 gen_s1 n_sub=200
18
+ [gen_s1] SDP+verify 88s plain ok 0.935; 16 failed-certified in 4 plants
19
+ [gen_s1] refine done 108s
20
+ [gen_s1] refined ok at (5, 10, 20, 40): [0.95, 0.95, 0.95, 0.983] (K=1) (115s)
21
+ seed9001 gen_s2 n_sub=200
22
+ [gen_s2] SDP+verify 91s plain ok 0.933; 16 failed-certified in 4 plants
23
+ [gen_s2] refine done 112s
24
+ [gen_s2] refined ok at (5, 10, 20, 40): [0.95, 0.967, 0.95, 0.983] (K=1) (118s)
results/furuta_v4/ddim_summary.txt ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ steps K=1: pooled | diff to DDPM-500 [95% CI] p K=4: pooled | diff to DDPM-500 [95% CI] p K=8: pooled | diff to DDPM-500 [95% CI] p K=16: pooled | diff to DDPM-500 [95% CI] p
2
+ DDPM500 92.4 93.2 93.8 94.2
3
+ DDIM5 90.6 | -1.9 [-4.3,+0.3] p=0.15 92.7 | -0.6 [-2.2,+0.9] p=0.56 93.7 | -0.1 [-0.9,+0.7] p=1.00 94.3 | +0.1 [-0.7,+0.9] p=1.00
4
+ DDIM10 91.7 | -0.7 [-2.7,+0.8] p=0.52 93.5 | +0.3 [-0.5,+1.1] p=0.69 93.9 | +0.1 [-0.5,+0.7] p=1.00 94.4 | +0.3 [-0.2,+0.9] p=0.50
5
+ DDIM20 91.9 | -0.6 [-1.9,+0.4] p=0.50 93.4 | +0.2 [-0.2,+0.6] p=0.75 93.9 | +0.1 [+0.0,+0.3] p=1.00 94.5 | +0.4 [+0.0,+1.0] p=0.50
6
+ DDIM50 91.9 | -0.6 [-1.5,+0.2] p=0.34 93.2 | +0.0 [-0.6,+0.6] p=1.00 93.5 | -0.3 [-0.9,+0.2] p=0.75 94.4 | +0.3 [-0.3,+1.1] p=1.00
results/furuta_v4/ddim_unetA_ema999_10.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ unetA_ema999 DDIM 10
2
+ K=1: 91.4% K=4: 93.1% K=8: 93.3% K=16: 94.2%
results/furuta_v4/ddim_unetA_ema999_5.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ unetA_ema999 DDIM 5
2
+ K=1: 90.0% K=4: 92.5% K=8: 93.9% K=16: 94.7%
results/furuta_v4/ddim_unetA_ema999_50.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ unetA_ema999 DDIM 50
2
+ K=1: 91.7% K=4: 93.1% K=8: 93.1% K=16: 94.7%
results/furuta_v4/ddim_unetAseed1_ema999_50.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ unetAseed1_ema999 DDIM 50
2
+ K=1: 92.2% K=4: 93.3% K=8: 93.3% K=16: 93.9%
results/furuta_v4/ddim_unetAseed2_ema999_10.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ unetAseed2_ema999 DDIM 10
2
+ K=1: 91.9% K=4: 93.9% K=8: 94.4% K=16: 95.0%
results/furuta_v4/ddim_unetAseed2_ema999_20.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ unetAseed2_ema999 DDIM 20
2
+ K=1: 91.7% K=4: 93.3% K=8: 94.4% K=16: 95.0%
results/furuta_v4/ddim_unetAseed2_ema999_5.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ unetAseed2_ema999 DDIM 5
2
+ K=1: 91.4% K=4: 93.1% K=8: 93.9% K=16: 94.2%
results/furuta_v4/ddim_unetAseed2_ema999_50.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ unetAseed2_ema999 DDIM 50
2
+ K=1: 91.7% K=4: 93.3% K=8: 94.2% K=16: 94.7%
results/furuta_v4/deploy_sidebyside.txt ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ deployment mismatch, K=8, nominal-conditioned generated+P, mean over 3 sampler seeds, n=120 plants; last columns: McNemar (retr-only - model-only, p) of each v4 model's seed 0 vs retrieval
2
+ scenario retr NMPC v3e2400 v4unetA v4unetS vs retr v4unetA vs retr v4unetS
3
+ nominal 90.0% 95.8% 91.4% 93.6% 95.8% 1-6 p=0.1250 0-7 p=0.0156
4
+ fric0.6 86.7% 95.8% 91.1% 93.6% 95.8% 1-10 p=0.0117 0-11 p=0.0010
5
+ fric1.6 9.2% 64.2% 10.3% 12.8% 10.8% 2-6 p=0.2891 1-4 p=0.3750
6
+ fric2.5 0.0% 29.2% 0.0% 0.0% 0.0% 0-0 p=1.0000 0-0 p=1.0000
7
+ ml_mp-_Lp- 68.3% 94.2% 70.8% 80.0% 76.1% 0-14 p=0.0001 5-13 p=0.0963
8
+ ml_mp-_Lp+ 81.7% 95.8% 89.2% 93.6% 95.6% 1-16 p=0.0003 1-17 p=0.0001
9
+ ml_mp+_Lp- 48.3% 95.0% 51.4% 65.6% 59.7% 4-25 p=0.0001 9-25 p=0.0090
10
+ ml_mp+_Lp+ 75.8% 95.0% 82.8% 91.9% 91.9% 2-22 p=0.0000 5-23 p=0.0009
11
+ both_f1.6_mp-_Lp- 9.2% 55.0% 11.9% 18.3% 14.7% 0-11 p=0.0010 2-9 p=0.0654
12
+ both_f1.6_mp-_Lp+ 19.2% 75.0% 18.9% 23.6% 22.8% 0-6 p=0.0312 2-6 p=0.2891
13
+ both_f1.6_mp+_Lp- 0.8% 52.5% 0.6% 0.0% 0.3% 1-0 p=1.0000 1-0 p=1.0000
14
+ both_f1.6_mp+_Lp+ 1.7% 67.5% 2.2% 3.3% 2.2% 0-2 p=0.5000 1-0 p=1.0000
15
+ both_f2.5_mp-_Lp- 0.0% 21.7% 0.3% 0.0% 0.0% 0-0 p=1.0000 0-0 p=1.0000
16
+ both_f2.5_mp-_Lp+ 0.0% 33.3% 0.0% 0.8% 0.3% 0-1 p=1.0000 0-0 p=1.0000
17
+ both_f2.5_mp+_Lp- 0.0% 18.3% 0.0% 0.6% 0.0% 0-0 p=1.0000 0-0 p=1.0000
18
+ both_f2.5_mp+_Lp+ 0.0% 33.3% 0.0% 0.0% 0.0% 0-0 p=1.0000 0-0 p=1.0000
19
+ max over scenarios of (model mean - retrieval mean), points: v3e2400: +7.5, v4unetA: +17.2, v4unetS: +16.1
results/furuta_v4/disjointness.txt ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ min over (eval plant, bank row) of sup-norm distance in (mp/0.024, Lp/0.129, log Dr, log Dp); a shared plant would give 0; typical bank NN spacing for scale:
2
+ bank LOO median 0.0695
3
+ draw 4242 bank3000 min 0.0145 median NN 0.0761
4
+ draw 4242 extra3000 min 0.0214 median NN 0.0681
5
+ draw 9001 bank3000 min 0.0276 median NN 0.0681
6
+ draw 9001 extra3000 min 0.0150 median NN 0.0696
7
+ draw 777 bank3000 min 0.0197 median NN 0.0711
8
+ draw 777 extra3000 min 0.0196 median NN 0.0655
results/furuta_v4/gate_mlp192k_ema999.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ mlp192k_ema999 ema999 models/v4/mlp_192k_s192000.pth
2
+ K=1: generated+P 66.7% nearest+P 54.2% K=4: generated+P 87.8% nearest+P 82.5% K=8: generated+P 89.7% nearest+P 90.0% K=16: generated+P 92.5% nearest+P 93.3%
results/furuta_v4/gate_mlp384k_ema9995.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ mlp384k_ema9995 ema9995 models/v4/mlp_384k_s384000.pth
2
+ K=1: generated+P 73.1% nearest+P 54.2% K=4: generated+P 89.4% nearest+P 82.5% K=8: generated+P 92.8% nearest+P 90.0% K=16: generated+P 95.3% nearest+P 93.3%
results/furuta_v4/gate_mlpA_ema999.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ mlpA_ema999 ema999 models/v4/mlp_a_s96000.pth
2
+ K=1: generated+P 60.8% nearest+P 54.2% K=4: generated+P 87.8% nearest+P 82.5% K=8: generated+P 91.4% nearest+P 90.0% K=16: generated+P 92.8% nearest+P 93.3%
results/furuta_v4/gate_mlpA_ema9995.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ mlpA_ema9995 ema9995 models/v4/mlp_a_s96000.pth
2
+ K=1: generated+P 55.3% nearest+P 54.2% K=4: generated+P 87.2% nearest+P 82.5% K=8: generated+P 91.1% nearest+P 90.0% K=16: generated+P 93.1% nearest+P 93.3%
results/furuta_v4/gate_mlpA_raw.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ mlpA_raw raw models/v4/mlp_a_s96000.pth
2
+ K=1: generated+P 55.6% nearest+P 54.2% K=4: generated+P 87.2% nearest+P 82.5% K=8: generated+P 91.1% nearest+P 90.0% K=16: generated+P 93.1% nearest+P 93.3%