Bridging the Theory-Practice Gap
Professor Wenzhuo Zhou’s research lies in the intersection of machine learning, statistics and artificial intelligence, with a primary focus on reinforcement learning, deep representation learning, and foundational models, particularly large language models. “My team and I are interested in leveraging statistical and mathematical tools to understand the nature of learning algorithms,” he says. “Our ultimate goal is to close the theory-practice gap and develop efficient, reliable, trustworthy algorithms to meet the demands of different scientific communities.”

Understanding the Nature of Learning Algorithms
With the emergence of massive datasets, the deep and reinforcement learning models trained on these datasets have acquired surprising capabilities, such as those seen in large language models. However, the theoretical explanations of these models are often incomplete, posing challenges for truly understanding their mechanisms. “My students and I are motivated to uncover the universal laws — such as sample efficiency, learning mechanisms, and model generalization — that underpin these phenomena,” he says. “Moreover, we can refine these models by aligning them with human preferences, selecting the model’s desired responses and behavior from a wide range of knowledge and abilities, ultimately achieving artificial general intelligence.” His team has applied these techniques across various domains, including natural language processing, healthcare and finance.

Promoting Scientific Collaboration
“As a machine learning researcher, one of my primary goals is to collaborate with domain experts in solving real-world problems,” says Professor Zhou. His team has been working with practitioners in the fields of cancer, diabetes and Alzheimer’s disease treatment and diagnosis, as well as in semantic search, video recommendation, financial management and robotic control. “In our collaborations with these investigators, we not only adapt existing methods but also develop new models and pipelines that are practical and robust in solving problems.”