Chao Du — AMI Labs

Biography

Chao Du is a Senior Research Scientist specializing in machine learning and deep learning, with a strong track record of advancing AI research across both academia and industry. Currently based in Singapore at Sea AI Lab (SAIL), he leads and contributes to cutting-edge research initiatives aimed at pushing the boundaries of scalable and applied artificial intelligence.

Prior to joining SAIL, Chao held research roles at Alibaba, where he worked on real-world AI applications, and completed research internships at NVIDIA and Microsoft Research Asia, gaining exposure to both industrial innovation and foundational research environments. His work spans core machine learning techniques, with a particular emphasis on deep learning and large-scale model development.

Chao earned his PhD in Machine Learning from Tsinghua University, one of the world’s leading institutions in computer science, where he also completed his undergraduate studies. Known for his strong technical foundation and practical mindset, he combines theoretical rigor with an ability to translate research into impactful applications. With expertise in Python and modern AI frameworks, Chao continues to contribute to the evolution of intelligent systems in high-impact domains.

Career History

2022-2026

Sea AI Lab (SAIL)

Research Scientist

2019-2022

Alibaba.com

Researcher

Key Papers

Nonparametric Data Attribution for Diffusion Models

Data attribution for generative models seeks to quantify the influence of individual training examples on model outputs. Existing methods for diffusion models typically require access to model gradients or retraining, limiting their applicability in proprietary or large-scale settings. We propose a nonparametric attribution method that operates entirely on data, measuring influence via patch-level similarity between generated and training images. Our approach is grounded in the analytical form of the optimal score function and naturally extends to multiscale representations, while remaining computationally efficient through convolution-based acceleration. In addition to producing spatially interpretable attributions, our framework uncovers patterns that reflect intrinsic relationships between training data and outputs, independent of any specific model. Experiments demonstrate that our method achieves strong attribution performance, closely matching gradient-based approaches and substantially outperforming existing nonparametric baselines.

2025 · Arxiv