summaryrefslogtreecommitdiff
AgeCommit message (Expand)Author
10 hoursAdd new experiment scripts, figures, and paper assets; untrack pyc/build arti...HEADmasterYurenHao0426
2026-04-26d=512 deep scan complete: FA+DFA at L=6,8,12 (10 seeds each)YurenHao0426
2026-04-26d=256 deep scan: FA+DFA at L=8 and L=12 (5 seeds each)YurenHao0426
2026-04-26CIFAR-100 per-seed diagnostics complete — full qualifying tableYurenHao0426
2026-04-26CIFAR-100 d=256 L=4: both FA and DFA fail — strongest qualifying settingYurenHao0426
2026-04-26BP+EP audit for d=512 L=2 qualifying seeds + CIFAR-100 supportYurenHao0426
2026-04-26Find setting where both FA and DFA fail: d=512 L=2 ResMLPYurenHao0426
2026-04-23Update NOTE.md + EVIDENCE_SUMMARY.md with FA results (2026-04-23)YurenHao0426
2026-04-22Add vanilla FA (Lillicrap 2016) implementation + full experiment suiteYurenHao0426
2026-04-08Add reproduce_all.ipynb: walkthrough for every paper figure + tableYurenHao0426
2026-04-08paper v2.39: restructure §1-§7 into subsection hierarchy (codex proposal)YurenHao0426
2026-04-08paper v2.38: ddof=1 statistical convention sweep (sample std with Bessel)YurenHao0426
2026-04-08paper v2.37.3: add paragraph headers to Appendix AYurenHao0426
2026-04-08paper v2.37.2: §7 ¶2 soften the reference-impl extensibility claimYurenHao0426
2026-04-08paper v2.37.1: abstract mentions nudging + training-loss confirmationYurenHao0426
2026-04-08paper v2.37: §7 add 'Open questions and concrete next experiments'YurenHao0426
2026-04-08paper v2.36.1: extend \paragraph{} headers to §1, §2, §7YurenHao0426
2026-04-08paper v2.36: add \paragraph{} headers throughout §3-§6 for readabilityYurenHao0426
2026-04-08paper v2.35: add Figure 2 - cross-method cos-vs-accuracy dissociationYurenHao0426
2026-04-08paper v2.34.3: fix Figure 4 cross-arch verdict matrix (data + layout)YurenHao0426
2026-04-08paper v2.34.2: §1 ¶3 DFA rescued deep cos +0.16 → +0.15YurenHao0426
2026-04-08Refresh paper/main.pdf after abstract-update revert experimentYurenHao0426
2026-04-08paper v2.34.1: SB/CB training loss decrease values from 3-seed (were s42)YurenHao0426
2026-04-08paper v2.34: §4 ¶4 mechanism hypothesis adds Mode 2 → Mode 1 causal chainYurenHao0426
2026-04-08paper v2.33: promote nudging test to §4 main text + correct to 3-seed valuesYurenHao0426
2026-04-08Sync EVIDENCE_SUMMARY.md and PAPER_OUTLINE.md with v2.32 valuesYurenHao0426
2026-04-08Sync experiment+protocol scripts with v2.32 corrected control valuesYurenHao0426
2026-04-08protocol/README.md: sync (c) range with v2.31.13 paper updateYurenHao0426
2026-04-08protocol/protocol.py: sync (c) range docstring with v2.31.13 paper updateYurenHao0426
2026-04-08paper v2.32.2: §3 ¶1 "20× below clamp" → "25× below clamp"YurenHao0426
2026-04-08paper v2.32.1: §5 ¶3 BP-to-SB gap 7.7 → 7.9 (cascade from BP+pen 0.530 â†...YurenHao0426
2026-04-08paper v2.32: BP+penalty multi-seeded (was single-seed s42)YurenHao0426
2026-04-08paper v2.31.14: §3 ¶1 local-loss formula sign-consistencyYurenHao0426
2026-04-08Commit dfa_pen_short lam=1e-4 s123/s456 JSONs (auditable source for §5 ¶2)YurenHao0426
2026-04-08Refresh paper/main.pdf after v2.31.13 sign-convention experiment revertYurenHao0426
2026-04-08paper v2.31.13: §6 ¶3 (c) ranges replaced with audit-data rangesYurenHao0426
2026-04-08Save threshold_sensitivity_output.txt for §6 ¶1 + Appendix E auditYurenHao0426
2026-04-08paper v2.31.12: §1 ¶2 + §1 ¶3 ‖g_L‖ values consistent with §3 ¶1YurenHao0426
2026-04-08Refresh paper/main.pdf to latest compiled stateYurenHao0426
2026-04-08Save null_calibration_penalized_dfa.json for §6 ¶2 auditYurenHao0426
2026-04-08paper v2.31.11: §3 ¶3 d=512 max-per-block growth uses sourced DFA valueYurenHao0426
2026-04-08paper v2.31.10: Appendix L drift values use 3-seed means (were s42)YurenHao0426
2026-04-08paper v2.31.9: relabel "StudentNet" → "no-terminal-LN ResMLP"YurenHao0426
2026-04-08paper v2.31.8: Appendix I EP random-target ‖h_L‖ values from saved JSONYurenHao0426
2026-04-08paper v2.31.7: Appendix H vanilla residual DFA endpoint valuesYurenHao0426
2026-04-08paper v2.31.6: §3 ¶2 no-residual values use 3-seed meansYurenHao0426
2026-04-08paper v2.31.5: §3 ¶3 ep-4 g_L from 3-seed (was single-seed g_2)YurenHao0426
2026-04-08paper v2.31.4: §2 ¶3 EP per-block growth 11.6× → 6.6× (3-seed)YurenHao0426
2026-04-08paper v2.31.3: §2 ¶3 per-block growth values were architecture mix-upYurenHao0426
2026-04-08paper v2.31.2: depth-scan layer-0 cos range [+0.39,+0.40] → [+0.38,+0.40]YurenHao0426
generated by cgit v1.2.3 (git 2.25.1) at 2026-06-14 19:05:31 +0000