Update README - GeoFractalDavid-Basin-k50 - Run 20251016_005644 - Acc 71.07%

README.md

@@ -23,7 +23,7 @@ model-index:
       type: imagenet-1k
     metrics:
     - type: accuracy
-      value: 70.45
+      value: 71.07
       name: Validation Accuracy
 ---
 
@@ -34,18 +34,18 @@ Features must "fit" geometric signatures: k-simplex shapes, Cantor positions, an
 
 ## 🎯 Performance
 
-- **Best Validation Accuracy**: 70.45%
-- **Epoch**: 6/10
-- **Training Time**: 13m
+- **Best Validation Accuracy**: 71.07%
+- **Epoch**: 8/10
+- **Training Time**: 18m
 
 ### Per-Scale Performance
-- **Scale 320D**: 62.87%
-- **Scale 384D**: 62.22%
-- **Scale 448D**: 59.92%
-- **Scale 512D**: 59.02%
-- **Scale 576D**: 66.66%
-- **Scale 640D**: 61.85%
-- **Scale 704D**: 55.60%
+- **Scale 320D**: 60.61%
+- **Scale 384D**: 59.73%
+- **Scale 448D**: 56.30%
+- **Scale 512D**: 55.22%
+- **Scale 576D**: 67.16%
+- **Scale 640D**: 62.33%
+- **Scale 704D**: 54.53%
 
 
 ## 🏗️ Architecture
@@ -73,8 +73,8 @@ Each scale learns to weight these components differently.
 The alpha parameter controls middle-interval weighting in the Cantor staircase.
 
 - **Initial**: 0.3290
-- **Final**: -0.0052
-- **Change**: -0.3342
+- **Final**: -0.0550
+- **Change**: -0.3840
 - **Converged to 0.5**: False
 
 The Cantor staircase uses soft triadic decomposition with learnable alpha to map
@@ -86,39 +86,39 @@ Each class has a learned scalar Cantor prototype. The model pulls features towar
 their class's Cantor position.
 
 **Scale 320D**:
-- Mean: 0.0413
-- Std: 0.0792
-- Range: [-0.1135, 0.2058]
+- Mean: 0.0267
+- Std: 0.0780
+- Range: [-0.1268, 0.1924]
 
 **Scale 384D**:
-- Mean: 0.0412
-- Std: 0.0792
-- Range: [-0.1137, 0.2056]
+- Mean: 0.0267
+- Std: 0.0780
+- Range: [-0.1267, 0.1921]
 
 **Scale 448D**:
-- Mean: 0.0412
-- Std: 0.0792
-- Range: [-0.1135, 0.2054]
+- Mean: 0.0267
+- Std: 0.0780
+- Range: [-0.1267, 0.1922]
 
 **Scale 512D**:
-- Mean: 0.0412
-- Std: 0.0792
-- Range: [-0.1134, 0.2054]
+- Mean: 0.0267
+- Std: 0.0780
+- Range: [-0.1265, 0.1924]
 
 **Scale 576D**:
-- Mean: 0.0417
-- Std: 0.0792
-- Range: [-0.1129, 0.2080]
+- Mean: 0.0267
+- Std: 0.0779
+- Range: [-0.1266, 0.1924]
 
 **Scale 640D**:
-- Mean: 0.0416
-- Std: 0.0792
-- Range: [-0.1129, 0.2078]
+- Mean: 0.0267
+- Std: 0.0780
+- Range: [-0.1268, 0.1925]
 
 **Scale 704D**:
-- Mean: 0.0415
-- Std: 0.0792
-- Range: [-0.1128, 0.2071]
+- Mean: 0.0267
+- Std: 0.0780
+- Range: [-0.1266, 0.1925]
 
 
 Most classes cluster around 0.5 (middle Cantor region), with smooth spread across [0,1].
@@ -128,13 +128,13 @@ This creates a continuous manifold rather than discrete bins.
 
 Each scale learns optimal weights for combining geometric components:
 
-**Scale 320D**: Feature=0.849, Cantor=0.033, Crystal=0.118
-**Scale 384D**: Feature=0.766, Cantor=0.035, Crystal=0.199
-**Scale 448D**: Feature=0.611, Cantor=0.037, Crystal=0.352
-**Scale 512D**: Feature=0.549, Cantor=0.037, Crystal=0.413
-**Scale 576D**: Feature=0.982, Cantor=0.002, Crystal=0.016
-**Scale 640D**: Feature=0.945, Cantor=0.003, Crystal=0.052
-**Scale 704D**: Feature=0.795, Cantor=0.004, Crystal=0.201
+**Scale 320D**: Feature=0.877, Cantor=0.026, Crystal=0.097
+**Scale 384D**: Feature=0.787, Cantor=0.028, Crystal=0.184
+**Scale 448D**: Feature=0.587, Cantor=0.030, Crystal=0.384
+**Scale 512D**: Feature=0.512, Cantor=0.030, Crystal=0.459
+**Scale 576D**: Feature=0.990, Cantor=0.001, Crystal=0.009
+**Scale 640D**: Feature=0.964, Cantor=0.002, Crystal=0.034
+**Scale 704D**: Feature=0.742, Cantor=0.003, Crystal=0.255
 
 
 **Pattern**: Lower scales rely on feature similarity, higher scales use crystal geometry.