XAI Dashboard

An interactive web application for visualizing and understanding deep learning model decisions through state-of-the-art explainability techniques.

Problem Statement

Deep learning models are often treated as “black boxes,” making it difficult to trust their predictions in critical applications like healthcare, autonomous vehicles, and financial systems. This dashboard bridges the gap by providing intuitive visualizations of model behavior.

Features

Explainability Methods

1. SHAP (SHapley Additive exPlanations)
2. Feature importance at instance and global levels
3. Force plots for individual predictions

4. Dependence plots for feature interactions

5. LIME (Local Interpretable Model-agnostic Explanations)

6. Local linear approximations of model behavior
7. Text and image explanations

8. Perturbation-based analysis

9. Grad-CAM (Gradient-weighted Class Activation Mapping)

10. Heatmap visualizations for CNNs
11. Layer-wise activation analysis

12. Multi-class attribution

13. Integrated Gradients

14. Path-based attribution method
15. Baseline comparison
16. Convergence analysis

Dashboard Capabilities

• Real-time Inference: Upload images or text for instant predictions
• Batch Analysis: Process multiple samples and compare explanations
• Model Comparison: Side-by-side comparison of different architectures
• Export Reports: Generate PDF/HTML reports of explanations
• Audit Trail: Log all predictions and explanations for compliance

Architecture

Frontend (React + TypeScript)
       ↓
API Gateway (FastAPI)
       ↓
┌──────────────┬──────────────┬──────────────┐
│  ML Models   │  Explainers  │   Storage    │
│  (PyTorch)   │  (SHAP/LIME) │ (PostgreSQL) │
└──────────────┴──────────────┴──────────────┘

Supported Models

• Image Classification (ResNet, EfficientNet, ViT)
• Object Detection (YOLO, Faster R-CNN)
• Text Classification (BERT, RoBERTa, GPT)
• Tabular Data (XGBoost, LightGBM, Neural Nets)

Installation

Backend

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Start API server
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Frontend

cd frontend
npm install
npm start

API Endpoints

Example Usage

import requests

# Upload image for classification
files = {'file': open('cat.jpg', 'rb')}
response = requests.post('http://localhost:8000/predict', files=files)
prediction = response.json()

# Get SHAP explanation
payload = {
    'model_id': 'resnet50',
    'image_id': prediction['image_id'],
    'method': 'deep_explainer'
}
response = requests.post('http://localhost:8000/explain/shap', json=payload)
explanation = response.json()

Use Cases

Healthcare

• Explaining disease diagnosis from medical images
• Identifying critical features in patient data
• Building trust with medical professionals

Autonomous Vehicles

• Understanding perception model decisions
• Debugging edge cases and failures
• Regulatory compliance and safety validation

Finance

• Credit risk assessment transparency
• Fraud detection explanation
• Regulatory reporting (GDPR, SR 11-7)

Performance

Measured on NVIDIA Tesla V100 GPU

Future Roadmap

• [ ] Add counterfactual explanations
• [ ] Implement attention-based explanations for transformers
• [ ] Multi-modal model support (e.g., CLIP)
• [ ] Automated fairness and bias detection
• [ ] Integration with MLOps platforms (MLflow, Weights & Biases)

Contributing

Contributions are welcome! Please read CONTRIBUTING.md for details on our code of conduct and the process for submitting pull requests.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Citation

If you use this dashboard in your research, please cite:

@software{kompas2024xai,
  author = {Kompas, Muhsin},
  title = {XAI Dashboard for Deep Learning},
  year = {2024},
  url = {https://github.com/muhsinkompas/xai-dashboard}
}

Acknowledgments

• SHAP library by Scott Lundberg
• LIME by Marco Tulio Ribeiro
• Grad-CAM implementation by PyTorch team