Update README - GeoFractalDavid-Basin-k50 - Run 20251016_011725 - Acc 67.78%

README.md

@@ -23,7 +23,7 @@ model-index:
       type: imagenet-1k
     metrics:
     - type: accuracy
-      value: 71.07
+      value: 67.78
       name: Validation Accuracy
 ---
 
@@ -34,18 +34,17 @@ Features must "fit" geometric signatures: k-simplex shapes, Cantor positions, an
 
 ## 🎯 Performance
 
-- **Best Validation Accuracy**: 71.07%
-- **Epoch**: 8/10
-- **Training Time**: 18m
+- **Best Validation Accuracy**: 67.78%
+- **Epoch**: 2/10
+- **Training Time**: 4m
 
 ### Per-Scale Performance
-- **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%
+- **Scale 448D**: 65.68%
+- **Scale 512D**: 65.72%
+- **Scale 576D**: 66.88%
+- **Scale 640D**: 65.49%
+- **Scale 704D**: 66.07%
+- **Scale 768D**: 65.25%
 
 
 ## 🏗️ Architecture
@@ -56,7 +55,7 @@ Features must "fit" geometric signatures: k-simplex shapes, Cantor positions, an
 - **Feature Dimension**: 512
 - **Number of Classes**: 1000
 - **k-Simplex Structure**: k=50 (51 vertices per class)
-- **Scales**: [320, 384, 448, 512, 576, 640, 704]
+- **Scales**: [448, 512, 576, 640, 704, 768]
 - **Total Simplex Vertices**: 51,000
 
 **Geometric Components**:
@@ -72,9 +71,9 @@ Each scale learns to weight these components differently.
 
 The alpha parameter controls middle-interval weighting in the Cantor staircase.
 
-- **Initial**: 0.3290
-- **Final**: -0.0550
-- **Change**: -0.3840
+- **Initial**: 0.3301
+- **Final**: 0.3377
+- **Change**: +0.0076
 - **Converged to 0.5**: False
 
 The Cantor staircase uses soft triadic decomposition with learnable alpha to map
@@ -85,40 +84,35 @@ features into [0,1] space with fractal structure.
 Each class has a learned scalar Cantor prototype. The model pulls features toward
 their class's Cantor position.
 
-**Scale 320D**:
-- Mean: 0.0267
-- Std: 0.0780
-- Range: [-0.1268, 0.1924]
-
-**Scale 384D**:
-- Mean: 0.0267
-- Std: 0.0780
-- Range: [-0.1267, 0.1921]
-
 **Scale 448D**:
-- Mean: 0.0267
-- Std: 0.0780
-- Range: [-0.1267, 0.1922]
+- Mean: 0.3299
+- Std: 0.1153
+- Range: [0.0698, 0.5232]
 
 **Scale 512D**:
-- Mean: 0.0267
-- Std: 0.0780
-- Range: [-0.1265, 0.1924]
+- Mean: 0.3303
+- Std: 0.1152
+- Range: [0.0707, 0.5232]
 
 **Scale 576D**:
-- Mean: 0.0267
-- Std: 0.0779
-- Range: [-0.1266, 0.1924]
+- Mean: 0.3406
+- Std: 0.1138
+- Range: [0.0846, 0.5392]
 
 **Scale 640D**:
-- Mean: 0.0267
-- Std: 0.0780
-- Range: [-0.1268, 0.1925]
+- Mean: 0.3284
+- Std: 0.1156
+- Range: [0.0675, 0.5210]
 
 **Scale 704D**:
-- Mean: 0.0267
-- Std: 0.0780
-- Range: [-0.1266, 0.1925]
+- Mean: 0.3376
+- Std: 0.1141
+- Range: [0.0799, 0.5346]
+
+**Scale 768D**:
+- Mean: 0.3321
+- Std: 0.1149
+- Range: [0.0728, 0.5256]
 
 
 Most classes cluster around 0.5 (middle Cantor region), with smooth spread across [0,1].
@@ -128,13 +122,12 @@ This creates a continuous manifold rather than discrete bins.
 
 Each scale learns optimal weights for combining geometric components:
 
-**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
+**Scale 448D**: Feature=0.653, Cantor=0.071, Crystal=0.276
+**Scale 512D**: Feature=0.610, Cantor=0.072, Crystal=0.318
+**Scale 576D**: Feature=0.879, Cantor=0.026, Crystal=0.096
+**Scale 640D**: Feature=0.578, Cantor=0.071, Crystal=0.351
+**Scale 704D**: Feature=0.822, Cantor=0.030, Crystal=0.148
+**Scale 768D**: Feature=0.668, Cantor=0.048, Crystal=0.285
 
 
 **Pattern**: Lower scales rely on feature similarity, higher scales use crystal geometry.