diff options
Diffstat (limited to 'PLAIN_CNN_CROSSOVER.md')
| -rw-r--r-- | PLAIN_CNN_CROSSOVER.md | 5 |
1 files changed, 4 insertions, 1 deletions
diff --git a/PLAIN_CNN_CROSSOVER.md b/PLAIN_CNN_CROSSOVER.md index 1687876..42c74b7 100644 --- a/PLAIN_CNN_CROSSOVER.md +++ b/PLAIN_CNN_CROSSOVER.md @@ -77,7 +77,10 @@ RMS soma-predictable traffic, one 64-example closed-form neutral predictor fit, and paired-neutral projection on every task minibatch. EP runs the complete LR grid `{0.003, 0.01, 0.03}` with beta 0.5, `dt=0.5`, 20 free and 4 nudged steps. BP, FA, DFA, clean-KP, SDIL, PEPITA, EP, and DP run 10 -global epochs in P1; FF runs 10 epochs per greedy layer. +global epochs in P1; FF runs 10 epochs per greedy layer. P1 keeps each +candidate rate constant for all 10 epochs (the PEPITA 60/90 boundaries are +therefore not reached), with no warmup. This makes the selector compare the +declared rates rather than different portions of a long-run decay schedule. For each gridded method select maximum best validation accuracy, then finite final loss, then the lower LR. If every candidate is nonfinite, retain and |