Speaker

Weishi Shi

Abstract

While machine learning and AI have achieved significant progress in many important domains, including computer vision and natural language processing, the lack of large-scale labeled data samples poses a grand challenge for the wide application of advanced statistical learning models in knowledge-rich domains, such as medicine, biology, physical science, and many more. Data annotation could be difficult, time-consuming, expensive, or even high-stakes in these domains. A standard way of training machine learning models may lead to a prohibitive cost of labeling the required data. So what is the hope of machine learning in knowledge-rich domains? In this talk, I will introduce active learning, an advanced interactive learning paradigm, as a viable solution to the data annotation crisis. I will demonstrate how my research augments the traditional models (e.g., support vector machines and neural networks) to allow them to learn effectively from less labeled data and how an intelligent learning machine could significantly save human annotation effort without sacrificing the predictive power. I will also discuss important extensions to support richer forms of human-machine collaborative learning as part of my future research plan.

Bio

Weishi Shi is a Ph.D. candidate from the Machine Learning and Data Intensive Computing research lab at Rochester Institute of Technology (RIT). His research interest lies in the general fields of machine learning and knowledge discovery, with a specific focus on active learning and its applications in diverse knowledge domains. His long-term research vision is to develop the next-generation machine learning paradigm, where the machine learns along with humans rather than blindly from humans. His research has been mainly published in top-tier machine learning and data mining venues, including NeurIPS, ICML, AISTATS, and ICDM. At RIT, Weishi has also taught multiple machine learning and data mining related courses, including Information Retrieval & Text Mining and Foundations of Data Mining.

Talk url | Passcode: 766398 [Video-off please during talk]