Deng Cai

Deng Cai is a computer scientist whose AI research connects feature selection, dimensionality reduction and clustering. He earns BEng and MEng degrees from Tsinghua University in 2000 and 2003, completes a computer science PhD at the University of Illinois Urbana-Champaign in 2009, and wins the David J. Kuck Outstanding PhD Thesis Award in 2010. His work includes the Laplacian Score, multi-cluster feature selection and landmark-based spectral clustering; in 2025, he coauthors STraj for trajectory prediction across geographic settings. As of October 2026, he is a professor and doctoral supervisor at Zhejiang University, where his research includes machine learning, data mining and computer vision.

Now at: Zhejiang University
Works: 429
Citations: 33,839
h-index: 77
Badges: Bridge
Boomerang
Wayfarer

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

2000–2003

Engineering degrees at Tsinghua

He earns a BEng and an MEng from Tsinghua University in 2000 and 2003, respectively.
2005

The Laplacian Score

With Xiaofei He and Partha Niyogi, he introduces Laplacian Score for Feature Selection at NIPS 2005. The filter method works in supervised or unsupervised settings and scores each feature by how well it preserves local structure.
2009

A computer science PhD

He completes a PhD in computer science at the University of Illinois Urbana-Champaign in 2009, with Jiawei Han as doctoral committee chair. His dissertation introduces spectral regression for dimensionality reduction, framing embedding as a regression problem that avoids eigen-decomposition of dense matrices.
2010

The Kuck thesis award

Illinois gives him the David J. Kuck Outstanding PhD Thesis Award in 2010.
2010

Multi-cluster feature selection

At KDD 2010, he coauthors Unsupervised Feature Selection for Multi-Cluster Data with Chiyuan Zhang and Xiaofei He.
2011

Landmark-based spectral clustering

With Xinlei Chen, he introduces landmark-based spectral clustering for large datasets at AAAI 2011. The method selects representative points as landmarks, represents other points as linear combinations of those landmarks, and scales linearly with the problem size.
2025

Cross-geography trajectory prediction

In 2025, he coauthors STraj, a self-training pipeline for trajectory prediction across geographic settings. It generates and updates pseudo trajectories, then uses high-consistency pseudo trajectories to supervise target-domain samples.
As of October 2026

Professor at Zhejiang University

He is a professor and doctoral supervisor in Zhejiang University's College of Computer Science and Technology, where his research includes machine learning, data mining and computer vision. In 2025, he is affiliated with the university's State Key Lab of CAD&CG.