Add EP to BP gradient sparsity analysis

EP CIFAR d=256: s(1e-6)=100%, mean_norm=1.41e-04
EP produces networks where ALL samples have non-zero BP gradients,
unlike DFA (0.4%), SB (21%), CB (3%). EP is closer to BP (98.7%).

Updated clean_sparsity_summary.csv (980 rows, now includes EP).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

1 files changed, 91 insertions, 0 deletions

diff --git a/results/confirmatory/clean_sparsity/cifar_ep_s789.json b/results/confirmatory/clean_sparsity/cifar_ep_s789.json
new file mode 100644
index 0000000..3aa00b4
--- /dev/null
+++ b/results/confirmatory/clean_sparsity/cifar_ep_s789.json
@@ -0,0 +1,91 @@
+{
+  "method": "ep",
+  "seed": 789,
+  "d": 256,
+  "per_layer": [
+    {
+      "layer": 0,
+      "mean_norm": 0.00017598697741050273,
+      "median_norm": 0.0001435329031664878,
+      "support": {
+        "1e-08": 1.0,
+        "1e-07": 1.0,
+        "1e-06": 1.0,
+        "1e-05": 1.0,
+        "0.0001": 0.77734375
+      },
+      "r_inf_mean": 0.25733304023742676,
+      "pr_mean": 0.21203935146331787,
+      "hoyer_mean": 0.5531952381134033,
+      "topk_energy": {
+        "1": 0.11893332749605179,
+        "5": 0.376035213470459,
+        "10": 0.5429649353027344,
+        "25": 0.7928071022033691
+      }
+    },
+    {
+      "layer": 1,
+      "mean_norm": 0.00017575427773408592,
+      "median_norm": 0.00014246668433770537,
+      "support": {
+        "1e-08": 1.0,
+        "1e-07": 1.0,
+        "1e-06": 1.0,
+        "1e-05": 1.0,
+        "0.0001": 0.77734375
+      },
+      "r_inf_mean": 0.2575938105583191,
+      "pr_mean": 0.21092239022254944,
+      "hoyer_mean": 0.5521162152290344,
+      "topk_energy": {
+        "1": 0.11935224384069443,
+        "5": 0.37730705738067627,
+        "10": 0.5443674325942993,
+        "25": 0.7936674952507019
+      }
+    },
+    {
+      "layer": 2,
+      "mean_norm": 0.00017053575720638037,
+      "median_norm": 0.00013913025031797588,
+      "support": {
+        "1e-08": 1.0,
+        "1e-07": 1.0,
+        "1e-06": 1.0,
+        "1e-05": 1.0,
+        "0.0001": 0.73828125
+      },
+      "r_inf_mean": 0.26289018988609314,
+      "pr_mean": 0.20421040058135986,
+      "hoyer_mean": 0.548264741897583,
+      "topk_energy": {
+        "1": 0.12429583817720413,
+        "5": 0.3861568570137024,
+        "10": 0.5497573614120483,
+        "25": 0.7960729598999023
+      }
+    },
+    {
+      "layer": 3,
+      "mean_norm": 0.00016748139751143754,
+      "median_norm": 0.00013177764776628464,
+      "support": {
+        "1e-08": 1.0,
+        "1e-07": 1.0,
+        "1e-06": 1.0,
+        "1e-05": 1.0,
+        "0.0001": 0.71484375
+      },
+      "r_inf_mean": 0.269894003868103,
+      "pr_mean": 0.19402557611465454,
+      "hoyer_mean": 0.5389708280563354,
+      "topk_energy": {
+        "1": 0.13044467568397522,
+        "5": 0.3994868993759155,
+        "10": 0.5619409680366516,
+        "25": 0.8028245568275452
+      }
+    }
+  ]
+}
\ No newline at end of file