Depth-utility ladder: trainable-block sweep (BP/FA/DFA) on ResMLP CIFAR-10HEADmaster

Appendix experiment triangulating the depth-utility diagnostic (D3) by varying
the number of trainable residual blocks k (last-k trainable, first L-k frozen at
init; embed/LN/head always trained).

- d=256 L=4 and d=512 L=2, 3 seeds, recipe identical to the main audit.
- BP climbs monotonically (+22-23pp); DFA peaks at the frozen baseline (k=0) and
  declines once any deep block is trained; FA shows partial/no net depth utility.
- Cross-checks reproduce existing anchors (BP 0.617, DFA 0.301, FA 0.402, frozen 0.349).
- frozen_init_identity_check quantifies frozen stack as a near-norm-preserving
  random feature map (per-block ||f||/||h||~0.10, stack cos 0.981), explaining the
  above-chance k=0 rung.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

1 files changed, 14 insertions, 0 deletions

diff --git a/run_depth_ladder.sh b/run_depth_ladder.sh
new file mode 100755
index 0000000..e450b47
--- /dev/null
+++ b/run_depth_ladder.sh
@@ -0,0 +1,14 @@
+#!/bin/bash
+set -e
+cd /home/yurenh2/fa
+echo "[$(date)] START primary d=256 L=4 ladder"
+CUDA_VISIBLE_DEVICES=2 python3 experiments/depth_utility_ladder.py \
+  --d_hidden 256 --num_blocks 4 --dataset cifar10 \
+  --methods bp fa dfa --k_values 0 1 2 3 4 --seeds 42 123 456 \
+  --epochs 100 --gpu 0 --output_dir results/depth_ladder
+echo "[$(date)] START secondary d=512 L=2 FA-failure ladder"
+CUDA_VISIBLE_DEVICES=2 python3 experiments/depth_utility_ladder.py \
+  --d_hidden 512 --num_blocks 2 --dataset cifar10 \
+  --methods bp fa dfa --k_values 0 1 2 --seeds 42 123 456 \
+  --epochs 100 --gpu 0 --output_dir results/depth_ladder
+echo "[$(date)] ALL DONE"