Xinwei Niu’s Personal Website

I am an MSc graduate in Artificial Intelligence from the University of Edinburgh, School of Informatics. Previously, I received Bachelor of Science with Honours in Artificial Intelligence at the University of Manchester, Department of Computer Science. During my undergraduate studies, I interned at the Language Understanding Lab of Samsung Research China, Beijing, where I was first exposed to multimodal large language models and their integration of vision–language representations.

My research interests span a broad range of efficient generative modelling techniques, including applications in both Natural Language Processing and image and video synthesis. I am particularly interested in sparse attention methods, including hardware-aware sparse attention, KV-cache eviction and compression methods. Mean while, I am interested in efficient generative architectures ranging from autoregressive and semi-autoregressive models to diffusion-based language models. I also follow advances in emerging state-space model and linear RNN architectures, especially in the context of building scalable, efficient and low-cost generative systems. I personally conducted a literature review on efficient diffusion models for image synthesis.

I am currently applying for PhD programs to continue my research in efficient, multimodal generative modelling and next-generation model architectures. If you would like to connect, collaborate, or simply chat about research, please feel free to reach out at nxw2002 [at] outlook [dot] com! I’m always happy to connect with new people and exchange ideas.

My CV is here.

During my spare time, I like to travel around different places, mainly in Europe.

Farewell to my little hammy, Yashi (ヤシ, 🥥).