Research Overview

My research revolves around Data-Centric AI, with a strong emphasis on improving the quality, reliability, and security of data used in machine learning and artificial intelligence systems. Key areas of interest include:

Deep Learning Data Optimization and Explainability

I am among the first to propose and systematically explore data optimization as a potential subfield or paradigm within machine learning. This research unifies various data-centric strategies, such as resampling, augmentation, perturbation, weighting, and pruning. I have developed new algorithms for deep learning data optimization and conducted rigorous studies on their theoretical underpinnings and practical effectiveness. Additionally, I investigate the explainability of deep learning models and components to enhance interpretability and trustworthiness.

Data Synthesis and Data Poisoning

To address limitations in current training data generation methods—such as insufficient diagnostic capabilities, limited focus on global data quality, and inadequate attention to security—I aim to construct a comprehensive framework for generating high-quality training data. This framework includes:

1. Multidimensional Data Perception
2. Training Set Diagnosis
3. Generation Strategy Design
4. Specific Sample Generation
5. Data Safety Evaluation and Defense

This research seeks to ensure that generated data is not only effective for learning tasks but also secure and reliable against adversarial threats.

Intelligent Analysis of Text and Image Data

I apply deep learning theories and methods to real-world challenges in both social and industrial domains. Specific applications include:

• Intelligent understanding of high-throughput internet text and image data
• Analysis and parsing of industrial control circuits
• Visual analysis for business intelligence

These methods have been successfully deployed in collaborations with over ten large and medium-sized enterprises, demonstrating the practical impact of my research.

Recruitment Opportunities

Postdoctoral Researchers

We are looking to recruit 1–2 postdoctoral researchers in the near future. Competitive benefits are provided, including a living/housing allowance of 400,000–1,400,000 RMB for those who choose to stay and work in Hangzhou after their postdoc (subject to relevant policies).

Graduate Students

• Ph.D. Students: Admission depends on annual quota allocation.
• Master’s Students: Approximately 2 positions are available each year.

Internships

We welcome graduate students with strong programming skills and a keen interest in AI data generation and security to join us for internships lasting 6 months or longer.

For inquiries, please contact: wuou@ucas.ac.cn