faeval.git/experiments/cnn_baseline.py, branch master

Fix CNN compute_bp_grads: remove inter-layer detach so gradients flow to all layers

2026-04-03T22:27:08+00:00

Old code detached hidden states between layers, making layers 0-2 disconnected
from the loss (gradient = None → 0). Fixed by keeping the forward graph connected.

BP CNN Gamma per-layer now: [0.985, 0.990, 0.987, 0.967] (was [0, 0, 0, 0.967])
But gradient norms are ~1e-17 (genuine numerical precision issue with CNN architecture).

Co-Authored-By: Claude Opus 4.6 (1M context)

Fix EP credit sign in cnn_baseline.py

2026-04-03T15:22:38+00:00

Co-Authored-By: Claude Opus 4.6 (1M context)

Fix CNN state bridge: use custom CNNStateBridge for variable input dims

2026-04-02T21:21:21+00:00

Co-Authored-By: Claude Opus 4.6 (1M context)

Add SB and CB methods to cnn_baseline.py

2026-04-02T21:19:14+00:00

State bridge: per-layer StateBridgeNet predicting h3 from flattened h_l
Credit bridge: per-layer ValueNet with terminal + bridge consistency + DFA warmup

Co-Authored-By: Claude Opus 4.6 (1M context)

Add CNN baseline: SmallCNN with BP/DFA/EP on CIFAR-10

2026-04-02T16:28:13+00:00

Co-Authored-By: Claude Opus 4.6 (1M context)