SoCSAE Training Visualization

Top-k sparse features and label-level overlap tracking

Optimization

Optimization Step0 / 1000

Data Shape

Concepts (n)6

Input Dim8

Expansion Term5

Top-k6

Points / Label10

Samples / Pair2

SoCSAE Methods

SupCon GeLU extra layer

Metrics

Batch (m*n)

Feature Dim

MSE Loss

0.0000

SupCon Loss

0.0000

Cluster Rate

0.0%