Javier Zazo
Senior Research Machine Learning Engineer
关于
I am a Senior Researcher at Microsoft Research Cambridge working on machine learning methods for efficient and scalable AI systems. Over the past two years, my research has focused on generative modelling, optimisation, and language-model architectures, including diffusion-based sequence modelling, multi-token prediction, and efficient large-scale training methods.
My recent work includes contributions to diffusion language models such as PUNT and Forward-Learned Discrete Diffusion Models, exploring alternatives to conventional autoregressive generation. I have also worked on Adaptive Rotations for Optimisation (ARO), investigating geometry-aware optimisation techniques for large-scale neural network training. More broadly, I am interested in developing architectures and training algorithms that improve model quality, efficiency, and scalability, with a particular focus on next-generation language modelling and generation paradigms.
Previously, I worked in Health Intelligence on the Antigen Map Project, in collaboration with Adaptive Biotechnologies. The project aimed to understand how T cell receptors (TCRs) bind to antigens and how our bodies protect us against infections, cancerous cells, and autoimmune diseases. My role focused on developing models and tools to identify which TCRs proliferate in response to disease, with a particular emphasis on sequence modelling and scalability to millions of TCRs.
I received my PhD from Universidad Politécnica de Madrid (UPM), where I worked on non-convex optimisation, game theory, and applications. I received my Telecommunications Engineering degree from UPM and Technische Universität Darmstadt (TUD). From 2018 to 2020, I was a postdoctoral fellow at Harvard University, where I worked on optimal transport, deep learning, and representation learning.