🧪 Interactive Research

ML Experiments Lab

Live visualization of CNN-Transformer Autoencoder · SHAP Explainability · Model Comparison

Interactive Mode

🔬 CNN-Transformer Output

Live Reconstruction

Score: — / undefined

×3
×0 Normal×7.5×15 Extreme

Status

✅ NORMAL

📊 Gradient-Based Explainability

SHAP Feature Attribution

Select Device:
Attribution = |∇_x Loss| × |x - x̂|

🧠 t-SNE Dimensionality Reduction

Latent Space Explorer

128D → 2D · Perplexity=30

⚔️ 4 Architectures Compared

Model Comparison Arena

87,813 datapoints

Training & Validation Loss

Multi-Metric Radar

🎯 Precision-Recall Tradeoff

Threshold Optimizer

Current: 1.3571

τ = 1.3571
0.2 (sensitive)↓ μ+3σ = 1.35714.0 (permissive)

Model

CNN-Transformer AE

Dataset

87,813 pts

Features

188

Devices

19

GPU

RTX 4070 Ti SUPER

ML Experiments Lab v1.0