About

I am a Principal Scientist and Head of the GenAI group at Samsung AI Center, Cambridge. My research focuses on building efficient, scalable, and robust generative AI systems for computer vision applications that enhances user-experience. I’m deeply committed to driving ideas end-to-end, from early exploration to peer-reviewed publications, and ultimately into real, deployable products used at scale. More broadly, my work aims to create generative technologies that are efficient, reliable, and ready for real-world deployment.

News

• We won the Samsung best paper (Silver) award 2025 for “On-Device High-Resolution Image Editing” project.

• New pre-print “FraQAT: Quantization Aware Training with Fractional bits” (https://arxiv.org/abs/2510.14823)

• Our paper “Edit: Efficient diffusion transformers with linear compressed attention” has been accepted at ICCV’25.

• Our paper “Upcycling Text-to-Image Diffusion Models for Multi-Task Capabilities” has been accepted at ICML’25.

• Our paper “Fast inference through the reuse of attention maps in diffusion models” has been accepted at CoRR’24.

• Paper accepted at ICLR’22: “Conditioning Sequence-to-sequence Models with Learned Activation Functions”.

• We won the Samsung best paper award 2021 for “Zero-cost Proxy Tasks for Light Weight NAS” project.

• Two papers accepted at ICLR’21: “NAS-Bench-ASR: Extending the Scope of NAS Algorithms to ASR Domain” and “Zero-cost Proxy Tasks for Light Weight NAS”.

• Our paper “Designing Robust Models for Behaviour Prediction using Sparse Data from Mobile Sensing: A Case Study of Office Workers” has been accepted at ACM HEALTH.

• Two papers accepted at INTERSPEECH’20: “Iterative Compression of End-to-End ASR Model using Reinforcement Learning” and “Bunched LPCNet: Vocoder for Low-cost Neural Text-To-Speech Systems.”.

• Our book “Intelligent Notification Systems” is now available as Synthesis Lectures on Mobile and Pervasive Computing (Morgan&Claypool Publishers) 2020.