matCercola18 commited on
Commit
27c122d
·
verified ·
1 Parent(s): c6dfe66

X28-M: held-out so3 confirmation, write-up

Browse files
docs/icml_r3_x28.md ADDED
@@ -0,0 +1,93 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ title: Generating the spline nonlinearity together with the controller (X28). SISO, no gain over phi = sat_1. MIMO, only odd splines keep a gain on new initial conditions; the study picks odd piecewise-linear knots at |x| = 1, 3, 5; the generator comparison is pending
3
+ date: 2026-11-13
4
+ status: SISO complete; MIMO optimum-level results complete (chosen class confirmed on held-out plants); MIMO generator comparison pending
5
+ claim: X28 (Matteo 2026-11-05): the July method (the diffusion model generates the 11-knot spline phi with the linear part) against the October setting (SISO phi = sat_1, X21; MIMO identity actuator, X27). If the spline loses, the measured loss documents why the setting changed.
6
+ maps_to: SISO section (why phi is fixed to sat_1), MIMO section (actuator class); no paper edit until Matteo agrees the claim
7
+ scripts: nonlinear/experiments/x28_siso.py (S1 gate, control, diagnostics, symmetric gate, S2 labels), nonlinear/training/train_x28.py (S3), nonlinear/experiments/x28_eval.py (S3 validation, S4 test), nonlinear/experiments/x28_paired.py (S4 paired test, fresh directions); nonlinear/mimo_unstable/x28_mimo.py (MIMO certificate, rollout, optimizer, gate, fresh directions, spline classes), nonlinear/mimo_unstable/x28_study_pick.py (class selection), nonlinear/mimo_unstable/x28_mimo_gen.py (MIMO S3/S4, pending)
8
+ artifacts: results/icml_r3/X28/{PREREG.md, gate_report.txt, control_report.txt, diag_report.txt, gate_sym_report.txt, labels_train/, labels_test/, eval_val_report.txt, val_clip_choice.json, eval_test/, eval_test_report.txt, eval_test_paired.txt, eval_test_fresh.txt}; results/icml_r3/X28/mimo/{selfcheck_m*.json, gate_*.txt, fresh_*.txt, study_m*/, study_choice.txt}; models/x28_{clS_s0,clS_s1,clS_s2,coefS_s0}/; HF matCercola18/DiffusionControl commits 268d7a3 (labels, models, gate), c6dfe66 (S4 test, study)
9
+ ---
10
+
11
+ # Generating the spline nonlinearity together with the controller (X28)
12
+
13
+ ## TL;DR
14
+ - **Question:** In July the diffusion model generated the nonlinearity phi (an 11-knot spline) together with the linear part. In October phi was fixed (SISO: sat_1; MIMO: identity). Does generating phi buy anything, and if not, what does it cost?
15
+ - **SISO answer:** No gain. At each plant's optimum the spline is cheaper than sat_1 by a median 2.84%, but all of it is knot asymmetry fitted to the 4 fixed initial directions of the cost: on 64 new directions it is no better (-0.36%), and an odd spline gains 0.00%. On the 199 test plants the spline generator ties the X21 sat_1 generator at 16 samples ("no measurable difference" on all 3 seed pairs), certifies 6 to 16 points fewer plants at 1 and 4 samples, and is 1.3% to 1.5% more expensive on new directions.
16
+ - **MIMO answer so far:** At the optimum every spline class beats identity by a median 4.5% to 6.6%, but only odd (sign-symmetric) classes keep a gain on new initial directions; free knots lose 2% to 6%. A pre-registered study over five classes picks odd piecewise-linear knots at |x| = 1, 3, 5 (3 numbers per channel). On 120 held-out plants per size it is a median 4.83% (m = 2) and 4.27% (m = 3) cheaper than identity at the optimum, and +2.96% and +0.94% on new initial directions, with no non-convergent rollout. The generator comparison is running.
17
+ - **So what:** The October choice phi = sat_1 is supported on SISO by a measured null result. On MIMO a small, odd actuator shape is worth generating, if the generator comparison confirms it.
18
+
19
+ ## 1. Problem / Goal
20
+ - **SISO setting (X21):** open-loop-unstable plants, bounded actuator, controller u = -phi(clip(F y, +-5)) with a dynamic filter F. X21 fixed phi = sat_1 and generated F. X28 generates F and the 10 nonzero knots of phi (|phi| <= 1, phi(0) = 0) with the same recipe.
21
+ - **MIMO setting (X27 family):** u = -Phi(K y), Phi = diag(psi_i), each psi_i a spline with knot ratios r = psi(x)/x in [0.1, 10], extended past +-5 along the ray, so the sector of each channel is exactly [min r, max r] and the circle criterion certifies the loop globally.
22
+ - **Falsification (PREREG):** the spline is "better" only if the paired median regret improves by at least 1 point with Holm p < 0.05 and the certified rate drops by at most 2 points, on at least 2 of 3 seed pairs; "worse" symmetrically; otherwise "no measurable difference".
23
+
24
+ ## 2. Method
25
+ - **S1 gate (per-plant optimum, 40 validation plants):** projected Adam on F and the knots from the X20b sat_1 label, gauge starts c = 1, 2, 5 (sat(y/c) with F -> cF is the same controller), cheapest snapshot certified by the regional circle criterion at s1. A control arm with frozen knots and the same optimizer and budget separates the spline from extra optimization of F. Diagnostics: odd (symmetrized) knots, and 64 new unit initial directions per plant (seed (7474, i)).
26
+ - **S2 labels:** each X21 training label refined jointly at c = 1 (1,971 training plants, 30,655 rows, 22,954 with knots off sat_1; training log of `x28_clS_s0`). Test plants also get c = 2, 5; these enter J* only.
27
+ - **S3 generators:** X21 recipe (DDPM-500, MLP 512 x 8, EMA 0.999, cost-slot conditioning). clS: closed-loop coordinates of F at the linearized gain k = (phi(1) - phi(-1))/2 plus 10 knots, seeds 0 to 2; coefS: slog of the coefficients plus knots, seed 0. Sampler clip and t-repair fixed on validation (addendum 2026-11-13): t-repair on every arm, X21 clip rule (lowest regret among clips certifying all 40 validation plants).
28
+ - **S4 test:** 199 X21 test plants, 16 samples per plant (64 for seed 0 as a secondary), J* = min of all F-only and joint labels, regret with uncertified = 1e3, paired against X21 clL scored by the same code.
29
+ - **MIMO gate:** 120 held-out plants per m (m = 2, 3), identity and spline arms from the same bank gain and the same Adam budget (400 steps, K lr 0.02 chosen on 64 validation bank plants), cost over the 2m positive unit directions (20 s, RK4 dt 0.01), circle LMI per channel (self-check against `family_sofm.verify`: 200 gains per m, 0 disagreements, `selfcheck_m*.json`). New directions: 64 random unit directions in R^(2m), both signs, seed (7575, id).
30
+ - **MIMO spline-class study (Matteo 2026-11-13):** five classes on the 64 validation bank plants per m: piecewise linear free (`sp`, 10 per channel), odd (`so`, 5), odd with knots at |x| = 1, 3, 5 (`so3`, 3), and cubic PCHIP on the ratios, free (`cf`, 10) and odd (`co`, 5); the PCHIP interpolant stays between its end values on each interval, so the sector stays exact. Rule filed before the run: eligible iff the new-direction gain over identity has CI lower bound > 0 at both m and at most 1% more non-convergent rollouts than identity; winner by mean gain, ties within 0.25 points to fewer parameters.
31
+
32
+ ## 3. Results
33
+
34
+ ### 3.1 SISO, per-plant optimum (40 validation plants)
35
+ | quantity | median [95% CI] | file |
36
+ |---|---|---|
37
+ | spline vs sat_1, gate d | 2.84% [1.06, 5.61], >= 2% on 23 of 40 | `gate_report.txt` |
38
+ | same optimizer and budget, knots frozen (control e) | 2.82% [1.01, 4.73] | `control_report.txt` |
39
+ | extra F optimization alone | 0.00% [0.00, 0.03] | `control_report.txt` |
40
+ | spline with its knots made odd, vs the spline | +3.85% [1.17, 5.78] dearer | `diag_report.txt` |
41
+ | spline vs frozen knots on 64 new directions | -0.36% [-4.87, 0.00] | `diag_report.txt` |
42
+ | odd-knot spline vs frozen knots (gate) | 0.00% [0.00, 0.00], >= 2% on 0 of 40 | `gate_sym_report.txt` |
43
+
44
+ The gain at the optimum is real for the cost as defined, and it is not extra F optimization. It comes from asymmetric knots that fit the 4 fixed initial directions; removing the asymmetry removes it, and it does not carry over to new directions. 3 of 40 best spline controllers did not re-certify in a fresh process (the certificate is order dependent); counting them as no gain gives e = 2.04% [0.43, 4.46] (PREREG addendum 2026-11-05).
45
+
46
+ ### 3.2 SISO, generators on the 199 test plants
47
+ | arm (t-repair, validation clip) | certified per sample | b = 1 certified | b = 16 certified | b = 16 regret vs J* |
48
+ |---|---|---|---|---|
49
+ | clS spline, seeds 0 / 1 / 2 (clip 1.5) | 52.2 / 52.1 / 52.7% | 49.2 / 52.8 / 49.7% | 95.0 / 96.5 / 95.5% | 7.14 / 6.33 / 6.30% |
50
+ | X21 clL sat_1, seeds 0 / 1 / 2 (clip 1.5) | 65.9 / 65.1 / 65.5% | 65.3 / 65.3 / 61.3% | 98.5 / 98.0 / 97.5% | 6.79 / 6.62 / 6.81% |
51
+ | coefS spline, seed 0 (clip 3) | 49.7% | 48.7% | 95.0% | 10.58% |
52
+
53
+ Source: `eval_test_report.txt`. Paired (seed j vs seed j, `eval_test_paired.txt`):
54
+ - **b = 16 (primary):** clL - clS median regret difference -0.04 [-0.61, +0.48], +0.19 [-0.42, +0.91], -0.32 [-1.19, +0.19] points, Holm p = 1 on all three; certified -3.5, -1.5, -2.0 points. **Verdict: no measurable difference.**
55
+ - **b = 1 and 4:** the spline generator certifies fewer plants: -16.1, -12.6, -11.6 points at b = 1 (McNemar p 0.00045 to 0.0095) and -6.0, -8.5, -6.5 at b = 4.
56
+ - **Without t-repair** (secondary) the spline generator loses at b = 16 by 0.75 to 2.92 points (Holm p <= 0.0014). coefS loses by 3.23 points.
57
+ - **64 samples, seed 0** (secondary): at b = 64 the spline generator is cheaper on the cost directions, 2.96% vs 4.07%, paired +0.44 [0.16, 0.91] points (Holm p 7.5e-05 within that block); at b = 16 in the same runs 6.61% vs 6.60%.
58
+ - **New directions** (`eval_test_fresh.txt`, best b = 16 sample per plant, 64 directions at s, against clL seed 0): clS -1.41% [-2.40, -0.30], -1.50% [-2.68, -0.30], -1.31% [-2.14, -0.30]; coefS -4.44%; clL seeds 1 and 2 +0.22% and +0.23% (seed-to-seed spread).
59
+ - **Knots:** the best b = 16 sample moves the knots from sat_1 by a median 0.077 to 0.086 and is not odd (max |phi(x) + phi(-x)| median 0.055 to 0.066).
60
+
61
+ ### 3.3 MIMO, per-plant optimum
62
+ Held-out gate, 120 plants per m (`gate_heldout_m*.txt`, `fresh_heldout_m*.txt`):
63
+ | class | m = 3 at optimum | m = 3 new directions | m = 2 at optimum | m = 2 new directions |
64
+ |---|---|---|---|---|
65
+ | free piecewise linear | 5.19% [4.55, 6.07] | -4.55% [-6.09, -3.55], 210 non-convergent | 6.36% [5.40, 7.05] | -1.58% [-2.64, -0.69], 131 non-convergent |
66
+ | odd piecewise linear | 4.36% [3.58, 5.14] | +0.54% [0.20, 1.27] | 5.52% [4.55, 6.36] | +2.71% [1.88, 3.39] |
67
+
68
+ Spline-class study, 64 validation plants per m (`gate_study_m*.txt`, `study_choice.txt`):
69
+ | class | parameters per channel | optimum m = 3 / m = 2 | new directions m = 3 | new directions m = 2 | eligible |
70
+ |---|---|---|---|---|---|
71
+ | `sp` free PWL | 10 | 6.31% / 6.61% | -5.94% | -2.11% | no |
72
+ | `so` odd PWL | 5 | 4.80% / 5.40% | +1.32% [0.64, 1.94] | +3.11% [2.48, 4.11] | yes |
73
+ | `so3` odd PWL, knots 1, 3, 5 | 3 | 4.54% / 5.22% | +1.82% [0.74, 2.57] | +3.99% [2.86, 5.13] | yes |
74
+ | `cf` free cubic | 10 | 5.39% / 6.39% | -6.06% | -3.09% | no |
75
+ | `co` odd cubic | 5 | 5.02% / 5.37% | +0.91% [0.45, 1.51] | +2.75% [2.13, 3.45] | yes |
76
+
77
+ **Choice: `so3`.** Interpolation (linear or cubic) does not matter; symmetry does. Fewer knots give the largest gain on new directions.
78
+
79
+ Held-out check of `so3` (120 plants per m, its own Adam call next to the stored arms, which reproduce exactly; `gate_heldout_m*.txt`, `fresh_heldout_m*.txt`):
80
+ | m | at optimum | new directions | non-convergent |
81
+ |---|---|---|---|
82
+ | 2 | 4.83% [4.14, 6.34], >= 2% on 97 of 120 | +2.96% [2.29, 3.65], > 0 on 106 of 120 | 0 |
83
+ | 3 | 4.27% [3.26, 4.80], >= 2% on 93 of 120 | +0.94% [0.53, 2.00], > 0 on 90 of 120 | 0 |
84
+
85
+ Both pre-registered rules hold at both m (gate: median >= 2% and CI lower bound > 0; new directions: CI lower bound > 0).
86
+
87
+ ### 3.4 MIMO, pending
88
+ - Generators (PREREG addendum and correction of 2026-11-13): labels on 7,936 (m = 3) and 3,936 (m = 2) bank plants, identity arm (a) against `so3` arm (b), 3 seeds each, 16 samples per held-out plant, the SISO verdict rule.
89
+
90
+ ## 4. Conclusion
91
+ - **SISO:** generating phi does not pay. The optimum-level gain is a fit to the cost's initial directions, the generator ties sat_1 at 16 samples, loses certified plants at small budgets, and does worse on new directions. This is the measured reason for the October setting phi = sat_1.
92
+ - **MIMO:** a free spline repeats the SISO artifact. An odd spline with 3 knots per channel keeps a gain on new initial directions at the optimum (held-out: +2.96% at m = 2, +0.94% at m = 3); whether a generator can deliver it is the pending comparison.
93
+ - **Caveats:** the SISO certificate is order dependent across calls (passes stand, fails prove nothing); the MIMO cost uses only positive unit directions, which is why the new-direction checks carry the conclusions; the 64-sample SISO result is one seed and was measured on the cost directions only.
results/icml_r3/X28/mimo/fresh_heldout_m2.npz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5c791032d09a9b00443dd57be45171d6fd4090ad964c4ba8d569020f1bca6f26
3
- size 589523
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:af32bbded30c32b2ec122fec52b8a3e59a56fc03a2b5039313bca14cc5080718
3
+ size 707381
results/icml_r3/X28/mimo/fresh_heldout_m2.txt CHANGED
@@ -1,3 +1,4 @@
1
- X28-M fresh directions, set heldout, m = 2: 64 random unit directions per plant
2
  sp: per plant median over directions of (J_id - J_sp) / J_sp; median over plants -1.58% [-2.64, -0.69], > 0 on 40 of 120; non-convergent rollouts id 0, sp 131
3
  so: per plant median over directions of (J_id - J_so) / J_so; median over plants 2.71% [1.88, 3.39], > 0 on 101 of 120; non-convergent rollouts id 0, so 1
 
 
1
+ X28-M fresh directions, set heldout, m = 2: 64 random unit directions per plant, seed (7575, plant id)
2
  sp: per plant median over directions of (J_id - J_sp) / J_sp; median over plants -1.58% [-2.64, -0.69], > 0 on 40 of 120; non-convergent rollouts id 0, sp 131
3
  so: per plant median over directions of (J_id - J_so) / J_so; median over plants 2.71% [1.88, 3.39], > 0 on 101 of 120; non-convergent rollouts id 0, so 1
4
+ so3: per plant median over directions of (J_id - J_so3) / J_so3; median over plants 2.96% [2.29, 3.65], > 0 on 106 of 120; non-convergent rollouts id 0, so3 0
results/icml_r3/X28/mimo/fresh_heldout_m3.npz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e77e9307eb3e35106dc12f56a3ee0c97f752cd6dcd50a1296655bee47873e2dd
3
- size 705844
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:62d2caa9900124afa63b38dad2ebcf74ed58743eab51cd45c783dd1339f699f9
3
+ size 823145
results/icml_r3/X28/mimo/fresh_heldout_m3.txt CHANGED
@@ -1,3 +1,4 @@
1
- X28-M fresh directions, set heldout, m = 3: 64 random unit directions per plant
2
  sp: per plant median over directions of (J_id - J_sp) / J_sp; median over plants -4.55% [-6.09, -3.55], > 0 on 17 of 120; non-convergent rollouts id 0, sp 210
3
  so: per plant median over directions of (J_id - J_so) / J_so; median over plants 0.54% [0.20, 1.27], > 0 on 84 of 120; non-convergent rollouts id 0, so 0
 
 
1
+ X28-M fresh directions, set heldout, m = 3: 64 random unit directions per plant, seed (7575, plant id)
2
  sp: per plant median over directions of (J_id - J_sp) / J_sp; median over plants -4.55% [-6.09, -3.55], > 0 on 17 of 120; non-convergent rollouts id 0, sp 210
3
  so: per plant median over directions of (J_id - J_so) / J_so; median over plants 0.54% [0.20, 1.27], > 0 on 84 of 120; non-convergent rollouts id 0, so 0
4
+ so3: per plant median over directions of (J_id - J_so3) / J_so3; median over plants 0.94% [0.53, 2.00], > 0 on 90 of 120; non-convergent rollouts id 0, so3 0
results/icml_r3/X28/mimo/gate_heldout_m2.txt CHANGED
@@ -1,12 +1,17 @@
1
  X28-M gate, set heldout, m = 2, n = 120, id certified on 120, median J_id 12.41
2
  id: snapshot 0 kept on 0, last snapshot on 20
3
- free knots (sp): d = (J_id - J_spline) / J_spline median 6.36% [5.40, 7.05], >= 2% on 109 of 120, mean 7.59% -> PAYS
4
  start sp_c1: certified 120, snapshot 0 kept on 1, median d 5.06%
5
  start sp_c2: certified 120, snapshot 0 kept on 1, median d 5.45%
6
  start sp_c5: certified 120, snapshot 0 kept on 3, median d 4.53%
7
  winning splines 120: max beta / alpha of a channel median 5.018; max |log r(-x) / r(x)| median 1.050; rollouts with final |x| > 1e-2: 0
8
- odd knots (so): d = (J_id - J_spline) / J_spline median 5.52% [4.55, 6.36], >= 2% on 99 of 120, mean 6.06% -> PAYS
9
  start so_c1: certified 120, snapshot 0 kept on 0, median d 4.14%
10
  start so_c2: certified 120, snapshot 0 kept on 0, median d 4.47%
11
  start so_c5: certified 120, snapshot 0 kept on 1, median d 3.48%
12
  winning splines 120: max beta / alpha of a channel median 3.873; max |log r(-x) / r(x)| median 0.000; rollouts with final |x| > 1e-2: 0
 
 
 
 
 
 
1
  X28-M gate, set heldout, m = 2, n = 120, id certified on 120, median J_id 12.41
2
  id: snapshot 0 kept on 0, last snapshot on 20
3
+ piecewise linear, free (sp, 10 per channel): d = (J_id - J_spline) / J_spline median 6.36% [5.40, 7.05], >= 2% on 109 of 120, mean 7.59% -> PAYS
4
  start sp_c1: certified 120, snapshot 0 kept on 1, median d 5.06%
5
  start sp_c2: certified 120, snapshot 0 kept on 1, median d 5.45%
6
  start sp_c5: certified 120, snapshot 0 kept on 3, median d 4.53%
7
  winning splines 120: max beta / alpha of a channel median 5.018; max |log r(-x) / r(x)| median 1.050; rollouts with final |x| > 1e-2: 0
8
+ piecewise linear, odd (so, 5 per channel): d = (J_id - J_spline) / J_spline median 5.52% [4.55, 6.36], >= 2% on 99 of 120, mean 6.06% -> PAYS
9
  start so_c1: certified 120, snapshot 0 kept on 0, median d 4.14%
10
  start so_c2: certified 120, snapshot 0 kept on 0, median d 4.47%
11
  start so_c5: certified 120, snapshot 0 kept on 1, median d 3.48%
12
  winning splines 120: max beta / alpha of a channel median 3.873; max |log r(-x) / r(x)| median 0.000; rollouts with final |x| > 1e-2: 0
13
+ piecewise linear, odd, knots at |x| = 1, 3, 5 (so3, 3 per channel): d = (J_id - J_spline) / J_spline median 4.83% [4.14, 6.34], >= 2% on 97 of 120, mean 5.72% -> PAYS
14
+ start so3_c1: certified 120, snapshot 0 kept on 0, median d 4.02%
15
+ start so3_c2: certified 120, snapshot 0 kept on 0, median d 4.58%
16
+ start so3_c5: certified 120, snapshot 0 kept on 1, median d 3.75%
17
+ winning splines 120: max beta / alpha of a channel median 3.825; max |log r(-x) / r(x)| median 0.000; rollouts with final |x| > 1e-2: 0
results/icml_r3/X28/mimo/gate_heldout_m3.txt CHANGED
@@ -1,12 +1,17 @@
1
  X28-M gate, set heldout, m = 3, n = 120, id certified on 120, median J_id 14.02
2
  id: snapshot 0 kept on 0, last snapshot on 23
3
- free knots (sp): d = (J_id - J_spline) / J_spline median 5.19% [4.55, 6.07], >= 2% on 99 of 120, mean 5.52% -> PAYS
4
  start sp_c1: certified 120, snapshot 0 kept on 1, median d 4.13%
5
  start sp_c2: certified 120, snapshot 0 kept on 2, median d 4.92%
6
  start sp_c5: certified 120, snapshot 0 kept on 1, median d 3.92%
7
  winning splines 119: max beta / alpha of a channel median 4.826; max |log r(-x) / r(x)| median 1.290; rollouts with final |x| > 1e-2: 0
8
- odd knots (so): d = (J_id - J_spline) / J_spline median 4.36% [3.58, 5.14], >= 2% on 98 of 120, mean 4.85% -> PAYS
9
  start so_c1: certified 120, snapshot 0 kept on 0, median d 3.68%
10
  start so_c2: certified 120, snapshot 0 kept on 1, median d 3.98%
11
  start so_c5: certified 120, snapshot 0 kept on 0, median d 2.26%
12
  winning splines 119: max beta / alpha of a channel median 4.464; max |log r(-x) / r(x)| median 0.000; rollouts with final |x| > 1e-2: 0
 
 
 
 
 
 
1
  X28-M gate, set heldout, m = 3, n = 120, id certified on 120, median J_id 14.02
2
  id: snapshot 0 kept on 0, last snapshot on 23
3
+ piecewise linear, free (sp, 10 per channel): d = (J_id - J_spline) / J_spline median 5.19% [4.55, 6.07], >= 2% on 99 of 120, mean 5.52% -> PAYS
4
  start sp_c1: certified 120, snapshot 0 kept on 1, median d 4.13%
5
  start sp_c2: certified 120, snapshot 0 kept on 2, median d 4.92%
6
  start sp_c5: certified 120, snapshot 0 kept on 1, median d 3.92%
7
  winning splines 119: max beta / alpha of a channel median 4.826; max |log r(-x) / r(x)| median 1.290; rollouts with final |x| > 1e-2: 0
8
+ piecewise linear, odd (so, 5 per channel): d = (J_id - J_spline) / J_spline median 4.36% [3.58, 5.14], >= 2% on 98 of 120, mean 4.85% -> PAYS
9
  start so_c1: certified 120, snapshot 0 kept on 0, median d 3.68%
10
  start so_c2: certified 120, snapshot 0 kept on 1, median d 3.98%
11
  start so_c5: certified 120, snapshot 0 kept on 0, median d 2.26%
12
  winning splines 119: max beta / alpha of a channel median 4.464; max |log r(-x) / r(x)| median 0.000; rollouts with final |x| > 1e-2: 0
13
+ piecewise linear, odd, knots at |x| = 1, 3, 5 (so3, 3 per channel): d = (J_id - J_spline) / J_spline median 4.27% [3.26, 4.80], >= 2% on 93 of 120, mean 4.61% -> PAYS
14
+ start so3_c1: certified 120, snapshot 0 kept on 1, median d 3.30%
15
+ start so3_c2: certified 120, snapshot 0 kept on 1, median d 3.67%
16
+ start so3_c5: certified 120, snapshot 0 kept on 0, median d 2.41%
17
+ winning splines 119: max beta / alpha of a channel median 3.993; max |log r(-x) / r(x)| median 0.000; rollouts with final |x| > 1e-2: 0
results/icml_r3/X28/mimo/heldout_m2/cert_so3_c1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d6df9af506bc218be057f9adcee3a862614d61eac6ac163a4d5764b5df63b99
3
+ size 18165
results/icml_r3/X28/mimo/heldout_m2/cert_so3_c2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b7b6b33886fb2f9de40197f337679c7ddeb8fb2bada12e42722dc56b5a030cf3
3
+ size 18026
results/icml_r3/X28/mimo/heldout_m2/cert_so3_c5.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1c57dafbf32150111a463ec20ae09498506ab034abd8f14757488dc8d97f2867
3
+ size 16950
results/icml_r3/X28/mimo/heldout_m2/so3_c1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e22a926397aa538edd2e73557ca11c448d440a272319b70f6ba8c16bd728d534
3
+ size 278475
results/icml_r3/X28/mimo/heldout_m2/so3_c2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7ad4fb95e20c4acde187b93add0db62c79cffade24499c821428e7d597b2d27e
3
+ size 275618
results/icml_r3/X28/mimo/heldout_m2/so3_c5.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3eb3408fb21bf21d851caacac7c0a3e23472a9144c61c004799290f8c1967c4c
3
+ size 256352
results/icml_r3/X28/mimo/heldout_m3/cert_so3_c1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dfc69b66f9ab8d5ae9027298b6e3a4eb8e6f1bc6f36cfee69d7307b374d70f88
3
+ size 28066
results/icml_r3/X28/mimo/heldout_m3/cert_so3_c2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4f02edf39266dcb6ebe4084a0a31039ec8a30058eb1658451411cf832e291d7b
3
+ size 27880
results/icml_r3/X28/mimo/heldout_m3/cert_so3_c5.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bcf5ccec0001308a4b82ac25fd2e09aa4a32b2a20fdf2535ff2920b0857afe94
3
+ size 26470
results/icml_r3/X28/mimo/heldout_m3/so3_c1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3c30275bccf1a478daca456c02b5feed8b5a24213c4ee54b4a638903aeb61566
3
+ size 454726
results/icml_r3/X28/mimo/heldout_m3/so3_c2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e19e9807730b41e3dd0b2ec595c922083815b115d2875b88c55ca7f6022d164
3
+ size 450535
results/icml_r3/X28/mimo/heldout_m3/so3_c5.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2ac180e131f747a7fd8332686123ae34d13ee025db4cb13eb300a8e51fd09f95
3
+ size 425705