Saqib Javed

I am a doctoral researcher at the CVLab at EPFL, working on Energy-Efficient Deep Networks under the supervision of Prof. Pascal Fua and Dr. Mathieu Salzmann. My current research focuses on 3D reconstruction using compressed Gaussian splatting, developing efficient quantized diffusion models, and reducing inference time for vision-language and large language models (VLMs and LLMs). These projects align with my broader goal of optimizing machine learning models for performance and efficiency.

Prior to my PhD, I worked on hardware-friendly mixed-precision neural networks under the supervision of Prof. Luca Benini at the Integrated Systems Laboratory, ETH Zurich. I hold a master’s degree from TU Munich and ETH Zurich, where I cultivated a strong interest in machine learning algorithms designed for low-power devices.

Email  /  CV  /  Twitter  /  Bluesky  /  Linkedin  /  Github

TC3DGS compresses dynamic 3D Gaussian representations using pruning, mixed-precision quantization, and trajectory interpolation, enabling efficient real-time applications.

SE-GS enhances novel view synthesis by reducing overfitting through uncertainty-aware perturbations and temporal regularization, outperforming existing methods.

QT-DoG leverages weight quantization to promote flatter minima, enhancing generalization across unseen domains while reducing model size and computational costs.

QT-DoG leverages weight quantization to promote flatter minima, enhancing generalization across unseen domains while reducing model size and computational costs.

Miscellanea