Dongdong Chen

I'm a Principal Research Manager at Microsoft GenAI in Washington, where I lead the foundational data science team. Our team is responsible for building a tightly integrated loop from data development to model training, playing a key role in improving the capabilities of OpenAI models in agentic coding scenarios. These efforts directly support agentic products such as GitHub Copilot and SWEAgent. In addition, I lead the post-training for Microsoft’s Phi family of small language models (SLMs), which are optimized for deployment on NPU devices in the latest Windows systems. I am the main model developer for many impactful Microsoft Product features, including but not limited to BackgroundRemoval, Generative Eraser and PPT summarization in Office Copilot.

My research interests includes large-scale data building for LLM development, multi-modality and single-modality pretraining \& post-training, deep generative models (e.g., GAN/Diffusion, Image-to-Image translation), general representation learning (such as fundamental network structure design), and AI security (e.g., adversarial learning and model IP protection).

I served as the Area Chairs of NeurIPs 23/24/25, CVPR 22/23, ECCV 2022, ICPR 22/24, WACV 24/25, SPC of AAAI 2022, and the Associated Editor of Pattern Recognition and IEEE TMM.

Email  /  CV  /  Google Scholar  

website template borrowed from source code. Thanks!