Upload from Ropedia Academy

README.md

@@ -48,6 +48,18 @@ Adam (lr 1e-3), 1200 steps, batch 128; 16 patches of 2×2, 50% masked; 2-layer t
 
 ![figure](figure.png)
 
+## Robustness (mean ± std over 5 seeds)
+
+Single-run numbers above are one seed; this is the distribution over independent re-trains (honest variance — no cherry-picking).
+
+
+| metric | mean ± std |
+|---|---|
+| `test_recon_mse` | 0.1349 ± 0.00074 |
+
+
+![seeds](seeds.png)
+
 ## Inference example
 
 ```python
@@ -86,6 +98,7 @@ jupyter nbconvert --to notebook --execute notebooks/training/B_mae_pretrain.ipyn
 - `figure.png`
 - `mae.pt`
 - `metrics.json`
+- `seeds.png`
 
 
 ## License

metrics.json

@@ -45,5 +45,19 @@
       0.1393
     ]
   ],
-  "test_recon_mse": 0.1364726573228836
+  "test_recon_mse": 0.1364726573228836,
+  "seeds": {
+    "n": 5,
+    "test_recon_mse": {
+      "mean": 0.1348876804113388,
+      "std": 0.0007431653355185816,
+      "values": [
+        0.1362,
+        0.1344,
+        0.134,
+        0.1347,
+        0.1351
+      ]
+    }
+  }
 }