Update README - GeoFractalDavid-Basin-k12 - Run 20251016_020120 - Acc 71.40%

README.md

@@ -23,7 +23,7 @@ model-index:
       type: imagenet-1k
     metrics:
     - type: accuracy
-      value: 71.13
+      value: 71.40
       name: Validation Accuracy
 ---
 
@@ -34,16 +34,16 @@ Features must "fit" geometric signatures: k-simplex shapes, Cantor positions, an
 
 ## 🎯 Performance
 
-- **Best Validation Accuracy**: 71.13%
-- **Epoch**: 8/10
-- **Training Time**: 14m
+- **Best Validation Accuracy**: 71.40%
+- **Epoch**: 10/10
+- **Training Time**: 18m
 
 ### Per-Scale Performance
-- **Scale 384D**: 61.51%
-- **Scale 512D**: 61.38%
-- **Scale 768D**: 70.29%
-- **Scale 1024D**: 51.67%
-- **Scale 1280D**: 44.19%
+- **Scale 384D**: 61.25%
+- **Scale 512D**: 60.67%
+- **Scale 768D**: 70.50%
+- **Scale 1024D**: 51.69%
+- **Scale 1280D**: 44.72%
 
 
 ## 🏗️ Architecture
@@ -71,8 +71,8 @@ Each scale learns to weight these components differently.
 The alpha parameter controls middle-interval weighting in the Cantor staircase.
 
 - **Initial**: 0.3290
-- **Final**: -0.0653
-- **Change**: -0.3944
+- **Final**: -0.0764
+- **Change**: -0.4055
 - **Converged to 0.5**: False
 
 The Cantor staircase uses soft triadic decomposition with learnable alpha to map
@@ -84,29 +84,29 @@ Each class has a learned scalar Cantor prototype. The model pulls features towar
 their class's Cantor position.
 
 **Scale 384D**:
-- Mean: 0.0252
-- Std: 0.0787
-- Range: [-0.1327, 0.1920]
+- Mean: 0.0226
+- Std: 0.0784
+- Range: [-0.1377, 0.1894]
 
 **Scale 512D**:
-- Mean: 0.0252
-- Std: 0.0787
-- Range: [-0.1330, 0.1924]
+- Mean: 0.0226
+- Std: 0.0784
+- Range: [-0.1377, 0.1895]
 
 **Scale 768D**:
-- Mean: 0.0253
-- Std: 0.0787
-- Range: [-0.1328, 0.1917]
+- Mean: 0.0227
+- Std: 0.0784
+- Range: [-0.1373, 0.1897]
 
 **Scale 1024D**:
-- Mean: 0.0253
-- Std: 0.0787
-- Range: [-0.1331, 0.1921]
+- Mean: 0.0226
+- Std: 0.0784
+- Range: [-0.1375, 0.1896]
 
 **Scale 1280D**:
-- Mean: 0.0253
-- Std: 0.0787
-- Range: [-0.1331, 0.1919]
+- Mean: 0.0227
+- Std: 0.0784
+- Range: [-0.1375, 0.1898]
 
 
 Most classes cluster around 0.5 (middle Cantor region), with smooth spread across [0,1].
@@ -116,11 +116,11 @@ This creates a continuous manifold rather than discrete bins.
 
 Each scale learns optimal weights for combining geometric components:
 
-**Scale 384D**: Feature=0.920, Cantor=0.022, Crystal=0.058
-**Scale 512D**: Feature=0.872, Cantor=0.026, Crystal=0.103
-**Scale 768D**: Feature=0.995, Cantor=0.001, Crystal=0.004
-**Scale 1024D**: Feature=0.942, Cantor=0.006, Crystal=0.052
-**Scale 1280D**: Feature=0.432, Cantor=0.003, Crystal=0.564
+**Scale 384D**: Feature=0.929, Cantor=0.020, Crystal=0.051
+**Scale 512D**: Feature=0.885, Cantor=0.023, Crystal=0.092
+**Scale 768D**: Feature=0.996, Cantor=0.001, Crystal=0.003
+**Scale 1024D**: Feature=0.952, Cantor=0.005, Crystal=0.043
+**Scale 1280D**: Feature=0.411, Cantor=0.003, Crystal=0.587
 
 
 **Pattern**: Lower scales rely on feature similarity, higher scales use crystal geometry.