summaryrefslogtreecommitdiff
path: root/results/confirmatory/A4_protocol_dependence.json
diff options
context:
space:
mode:
authorYurenHao0426 <Blackhao0426@gmail.com>2026-03-30 19:25:53 -0500
committerYurenHao0426 <Blackhao0426@gmail.com>2026-03-30 19:25:53 -0500
commit8b21fb32bf0997e3f4266c1c22414e49f1fdcfcc (patch)
tree54e3c678c8d45330c6085b02a27de82cc884e17d /results/confirmatory/A4_protocol_dependence.json
parent2a230acd5ee3fa6605892d524badf281ba7e9cfd (diff)
Add confirmatory paper experiments: A1-A4, all 10 seeds complete
A1: Synthetic nonlinearity ladder (240 rows: 3 alpha × 2 depth × 4 methods × 10 seeds) A2: CIFAR state-vs-credit counterexample (30 rows: 3 methods × 10 seeds) A3: Frozen vs online dissociation (60 rows: 2 regimes × 3 methods × 10 seeds) A4: Protocol dependence panel (82 rows: assembled from existing results) All experiments ran on GPU 3. Total runtime: ~20 hours. CSVs in results/confirmatory/. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Diffstat (limited to 'results/confirmatory/A4_protocol_dependence.json')
-rw-r--r--results/confirmatory/A4_protocol_dependence.json494
1 files changed, 494 insertions, 0 deletions
diff --git a/results/confirmatory/A4_protocol_dependence.json b/results/confirmatory/A4_protocol_dependence.json
new file mode 100644
index 0000000..59671d4
--- /dev/null
+++ b/results/confirmatory/A4_protocol_dependence.json
@@ -0,0 +1,494 @@
+[
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "dfa",
+ "metric": "gamma",
+ "value": 0.008470005544950254
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "dfa",
+ "metric": "rho",
+ "value": -0.022525018197484314
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "dfa",
+ "metric": "nudge",
+ "value": -2.3932622525535407e-05
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "dfa",
+ "metric": "loss_before",
+ "value": 1.1802499413490295
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "dfa",
+ "metric": "loss_after_1step",
+ "value": 1.1798909544944762
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "dfa",
+ "metric": "delta_loss_1step",
+ "value": -0.00035898685455326707
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "dfa",
+ "metric": "delta_acc_1step",
+ "value": 0.002343749999999978
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "scalar_cb",
+ "metric": "gamma",
+ "value": 0.12223921716213226
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "scalar_cb",
+ "metric": "rho",
+ "value": 0.09000759199261665
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "scalar_cb",
+ "metric": "nudge",
+ "value": -0.00036903251748299226
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "scalar_cb",
+ "metric": "loss_before",
+ "value": 1.1802499413490295
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "scalar_cb",
+ "metric": "loss_after_1step",
+ "value": 1.1835142850875855
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "scalar_cb",
+ "metric": "delta_loss_1step",
+ "value": 0.003264343738555997
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "scalar_cb",
+ "metric": "delta_acc_1step",
+ "value": 0.0007812499999999556
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "vec_eT_M4",
+ "metric": "gamma",
+ "value": 0.37760399281978607
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "vec_eT_M4",
+ "metric": "rho",
+ "value": 0.41141805797815323
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "vec_eT_M4",
+ "metric": "nudge",
+ "value": -0.0013705549063161016
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "vec_eT_M4",
+ "metric": "loss_before",
+ "value": 1.1802499413490295
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "vec_eT_M4",
+ "metric": "loss_after_1step",
+ "value": 1.1832772016525268
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "vec_eT_M4",
+ "metric": "delta_loss_1step",
+ "value": 0.0030272603034973145
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "vec_eT_M4",
+ "metric": "delta_acc_1step",
+ "value": -0.0031250000000000444
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "oracle_bp",
+ "metric": "gamma",
+ "value": 1.0
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "oracle_bp",
+ "metric": "rho",
+ "value": 0.9979714304208755
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "oracle_bp",
+ "metric": "nudge",
+ "value": -0.0031722619314678013
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "oracle_bp",
+ "metric": "loss_before",
+ "value": 1.1802499413490295
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "oracle_bp",
+ "metric": "loss_after_1step",
+ "value": 1.1796290874481201
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "oracle_bp",
+ "metric": "delta_loss_1step",
+ "value": -0.0006208539009093794
+ },
+ {
+ "slice": "snapshot_exploit_ep100",
+ "method": "oracle_bp",
+ "metric": "delta_acc_1step",
+ "value": 0.005468750000000022
+ },
+ {
+ "slice": "snapshot_ep5",
+ "method": "dfa",
+ "metric": "dl_held_1",
+ "value": 6.842613220214844e-05
+ },
+ {
+ "slice": "snapshot_ep5",
+ "method": "dfa",
+ "metric": "dl_same_1",
+ "value": -0.011249661445617676
+ },
+ {
+ "slice": "snapshot_ep5",
+ "method": "dfa",
+ "metric": "dl_held_5",
+ "value": 0.0032531023025512695
+ },
+ {
+ "slice": "snapshot_ep5",
+ "method": "dfa",
+ "metric": "dl_same_5",
+ "value": -0.004767775535583496
+ },
+ {
+ "slice": "snapshot_ep5",
+ "method": "vec_eT_M4",
+ "metric": "dl_held_1",
+ "value": -0.0002554655075073242
+ },
+ {
+ "slice": "snapshot_ep5",
+ "method": "vec_eT_M4",
+ "metric": "dl_same_1",
+ "value": -0.011568784713745117
+ },
+ {
+ "slice": "snapshot_ep5",
+ "method": "vec_eT_M4",
+ "metric": "dl_held_5",
+ "value": -0.005142092704772949
+ },
+ {
+ "slice": "snapshot_ep5",
+ "method": "vec_eT_M4",
+ "metric": "dl_same_5",
+ "value": -0.007385373115539551
+ },
+ {
+ "slice": "snapshot_ep5",
+ "method": "oracle_bp",
+ "metric": "dl_held_1",
+ "value": -0.0002244710922241211
+ },
+ {
+ "slice": "snapshot_ep5",
+ "method": "oracle_bp",
+ "metric": "dl_same_1",
+ "value": -0.011359930038452148
+ },
+ {
+ "slice": "snapshot_ep5",
+ "method": "oracle_bp",
+ "metric": "dl_held_5",
+ "value": -0.008973956108093262
+ },
+ {
+ "slice": "snapshot_ep5",
+ "method": "oracle_bp",
+ "metric": "dl_same_5",
+ "value": -0.008551597595214844
+ },
+ {
+ "slice": "snapshot_ep20",
+ "method": "dfa",
+ "metric": "dl_held_1",
+ "value": 0.0006439685821533203
+ },
+ {
+ "slice": "snapshot_ep20",
+ "method": "dfa",
+ "metric": "dl_same_1",
+ "value": -0.0050983428955078125
+ },
+ {
+ "slice": "snapshot_ep20",
+ "method": "dfa",
+ "metric": "dl_held_5",
+ "value": 0.0009417533874511719
+ },
+ {
+ "slice": "snapshot_ep20",
+ "method": "dfa",
+ "metric": "dl_same_5",
+ "value": -0.001807570457458496
+ },
+ {
+ "slice": "snapshot_ep20",
+ "method": "vec_eT_M4",
+ "metric": "dl_held_1",
+ "value": 4.982948303222656e-05
+ },
+ {
+ "slice": "snapshot_ep20",
+ "method": "vec_eT_M4",
+ "metric": "dl_same_1",
+ "value": -0.005342721939086914
+ },
+ {
+ "slice": "snapshot_ep20",
+ "method": "vec_eT_M4",
+ "metric": "dl_held_5",
+ "value": 0.0015401840209960938
+ },
+ {
+ "slice": "snapshot_ep20",
+ "method": "vec_eT_M4",
+ "metric": "dl_same_5",
+ "value": -0.00039839744567871094
+ },
+ {
+ "slice": "snapshot_ep20",
+ "method": "oracle_bp",
+ "metric": "dl_held_1",
+ "value": 6.651878356933594e-05
+ },
+ {
+ "slice": "snapshot_ep20",
+ "method": "oracle_bp",
+ "metric": "dl_same_1",
+ "value": -0.004968762397766113
+ },
+ {
+ "slice": "snapshot_ep20",
+ "method": "oracle_bp",
+ "metric": "dl_held_5",
+ "value": 0.0003821849822998047
+ },
+ {
+ "slice": "snapshot_ep20",
+ "method": "oracle_bp",
+ "metric": "dl_same_5",
+ "value": -0.0005745887756347656
+ },
+ {
+ "slice": "scaffold_3seed",
+ "method": "dfa",
+ "metric": "final_mean",
+ "value": 0.30593333333333333
+ },
+ {
+ "slice": "scaffold_3seed",
+ "method": "dfa",
+ "metric": "final_std",
+ "value": 0.006104825049818285
+ },
+ {
+ "slice": "scaffold_3seed",
+ "method": "dfa",
+ "metric": "acc20_mean",
+ "value": 0.30383333333333334
+ },
+ {
+ "slice": "scaffold_3seed",
+ "method": "dfa",
+ "metric": "acc20_std",
+ "value": 0.007825741001478518
+ },
+ {
+ "slice": "scaffold_3seed",
+ "method": "perlayer",
+ "metric": "final_mean",
+ "value": 0.30396666666666666
+ },
+ {
+ "slice": "scaffold_3seed",
+ "method": "perlayer",
+ "metric": "final_std",
+ "value": 0.006049977043115317
+ },
+ {
+ "slice": "scaffold_3seed",
+ "method": "perlayer",
+ "metric": "acc20_mean",
+ "value": 0.2874666666666667
+ },
+ {
+ "slice": "scaffold_3seed",
+ "method": "perlayer",
+ "metric": "acc20_std",
+ "value": 0.011382540235914942
+ },
+ {
+ "slice": "scaffold_3seed",
+ "method": "vec",
+ "metric": "final_mean",
+ "value": 0.3133333333333333
+ },
+ {
+ "slice": "scaffold_3seed",
+ "method": "vec",
+ "metric": "final_std",
+ "value": 0.007196912918436331
+ },
+ {
+ "slice": "scaffold_3seed",
+ "method": "vec",
+ "metric": "acc20_mean",
+ "value": 0.30333333333333334
+ },
+ {
+ "slice": "scaffold_3seed",
+ "method": "vec",
+ "metric": "acc20_std",
+ "value": 0.007024401912065005
+ },
+ {
+ "slice": "online_3seed",
+ "method": "dfa_s42",
+ "metric": "test_acc",
+ "value": 0.3116
+ },
+ {
+ "slice": "online_3seed",
+ "method": "dfa_s42",
+ "metric": "mean_gamma",
+ "value": 0.10076353600015864
+ },
+ {
+ "slice": "online_3seed",
+ "method": "dfa_s42",
+ "metric": "mean_rho",
+ "value": -0.00476757250726223
+ },
+ {
+ "slice": "online_3seed",
+ "method": "cb_eT_s42",
+ "metric": "test_acc",
+ "value": 0.283
+ },
+ {
+ "slice": "online_3seed",
+ "method": "cb_eT_s42",
+ "metric": "mean_gamma",
+ "value": 0.17921950668096542
+ },
+ {
+ "slice": "online_3seed",
+ "method": "cb_eT_s42",
+ "metric": "mean_rho",
+ "value": 0.008699589408934116
+ },
+ {
+ "slice": "online_3seed",
+ "method": "dfa_s123",
+ "metric": "test_acc",
+ "value": 0.3106
+ },
+ {
+ "slice": "online_3seed",
+ "method": "dfa_s123",
+ "metric": "mean_gamma",
+ "value": 0.11738517042249441
+ },
+ {
+ "slice": "online_3seed",
+ "method": "dfa_s123",
+ "metric": "mean_rho",
+ "value": 0.005885639227926731
+ },
+ {
+ "slice": "online_3seed",
+ "method": "cb_eT_s123",
+ "metric": "test_acc",
+ "value": 0.2753
+ },
+ {
+ "slice": "online_3seed",
+ "method": "cb_eT_s123",
+ "metric": "mean_gamma",
+ "value": 0.17617796920239925
+ },
+ {
+ "slice": "online_3seed",
+ "method": "cb_eT_s123",
+ "metric": "mean_rho",
+ "value": 0.0019302130676805973
+ },
+ {
+ "slice": "online_3seed",
+ "method": "dfa_s456",
+ "metric": "test_acc",
+ "value": 0.297
+ },
+ {
+ "slice": "online_3seed",
+ "method": "dfa_s456",
+ "metric": "mean_gamma",
+ "value": 0.09892195643624291
+ },
+ {
+ "slice": "online_3seed",
+ "method": "dfa_s456",
+ "metric": "mean_rho",
+ "value": -0.004888533148914576
+ },
+ {
+ "slice": "online_3seed",
+ "method": "cb_eT_s456",
+ "metric": "test_acc",
+ "value": 0.2893
+ },
+ {
+ "slice": "online_3seed",
+ "method": "cb_eT_s456",
+ "metric": "mean_gamma",
+ "value": 0.23417379707098007
+ },
+ {
+ "slice": "online_3seed",
+ "method": "cb_eT_s456",
+ "metric": "mean_rho",
+ "value": -0.002338360995054245
+ }
+] \ No newline at end of file