About This Project
Developed CNN architectures for facial detection and integrated a Debiasing Variational Autoencoder (DB-VAE) to reweight training samples. Improved fairness by increasing sampling for underrepresented facial features, reducing demographic bias in model predictions.
Key Features
- CNN-based facial detection
- Latent variable reweighting
- Fairness-aware training
- Bias visualization
