DeepLense
Ensemble DenseNet for gravitational lensing classification (96.4% accuracy) and VAE for synthetic astronomical image generation. GSoC 2025 Evaluation – ML4SCI
Google Summer of Code 2025 Evaluation – ML4SCI (Apr 2025)
- Designed ensemble DenseNet architecture achieving 96.4% accuracy in multi-class classification of gravitational lensing images; implemented custom ResNet for continuous parameter regression (RMSE 0.034).
- Built convolutional Variational Autoencoder (VAE) with 256-dimensional latent space for synthetic data generation; achieved 87% structural similarity (SSIM) between generated and real astronomical images.