From 88485f6f05243e2fc459807b6644c1ff45a55d2e Mon Sep 17 00:00:00 2001
From: YurenHao0426 <Blackhao0426@gmail.com>
Date: Fri, 3 Apr 2026 17:35:12 -0500
Subject: Recompute all CNN diagnostics with fixed BP gradient flow
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

CNN CIFAR-10 (5 seeds, fixed Gamma):
BP: acc=86.8%, Gamma=0.970, rho=0.603
DFA: acc=56.7%, Gamma=0.896, rho=0.061
EP: acc=50.6%, Gamma=0.484, rho=0.450
SB: acc=63.3%, Gamma=1.000 (BP self-cos, feedback nets not saved), rho=0.601
CB: acc=31.8%, Gamma=1.000 (BP self-cos), rho=0.226

DFA Gamma=0.896 is notably high — CNN DFA credit aligns well with BP gradients.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---
 results/cnn_baseline/ep_s123.json | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

(limited to 'results/cnn_baseline/ep_s123.json')

diff --git a/results/cnn_baseline/ep_s123.json b/results/cnn_baseline/ep_s123.json
index 5f06a86..f13a427 100644
--- a/results/cnn_baseline/ep_s123.json
+++ b/results/cnn_baseline/ep_s123.json
@@ -2,6 +2,18 @@
   "method": "ep",
   "seed": 123,
   "acc": 0.4897,
-  "Gamma": 0.24602489173412323,
-  "rho": 0.5164633095264435
+  "Gamma": 0.49209384838025017,
+  "rho": 0.47923217713832855,
+  "gammas_per_layer": [
+    2.8399340408213902e-06,
+    0.22918346524238586,
+    0.7567289471626282,
+    0.9824601411819458
+  ],
+  "rhos_per_layer": [
+    0.0,
+    0.0,
+    0.9310277700424194,
+    0.9859009385108948
+  ]
 }
\ No newline at end of file
-- 
cgit v1.2.3