diff options
| author | YurenHao0426 <Blackhao0426@gmail.com> | 2026-04-26 15:05:13 -0500 |
|---|---|---|
| committer | YurenHao0426 <Blackhao0426@gmail.com> | 2026-04-26 15:05:13 -0500 |
| commit | c4bd8d321b3cb2e77a6727ecfc2c2ae999065c4a (patch) | |
| tree | 5e14c147a758865071059c90fe379f04e782b694 /results/confirmatory/clean_sparsity/synth_dfa_s42_a1.0_L4.json | |
| parent | 1c172f7d038eedd3d828d453e852c060072f52c8 (diff) | |
d=256 deep scan: FA+DFA at L=8 and L=12 (5 seeds each)
FA does NOT drop below frozen 0.349 at deeper L on d=256:
L=8: FA mean 0.394, min 0.386 (gap +3.7pp)
L=12: FA mean 0.391, min 0.368 (gap +1.9pp)
FA accuracy is essentially depth-invariant (~0.39) even though FA
deep cosine drops from +0.13 (L=8) to +0.09 (L=12). DFA is always
below frozen (~0.27-0.30).
Conclusion: on CIFAR-10 with d=256 ResMLP, FA is too good at L≥4
to fail the frozen baseline. The only qualifying setting at deeper L
would require ~20+ seeds to find a rare 2σ outlier. The d=512 L=2
setting (seeds 1,2,5) remains the cleanest qualifying case.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Diffstat (limited to 'results/confirmatory/clean_sparsity/synth_dfa_s42_a1.0_L4.json')
0 files changed, 0 insertions, 0 deletions