Data Science Talk Series 

Title: From Past Lessons to Future Directions: Active Learning Is All You Need 

Abstract: Active learning is not just a machine learning technique: it is one of the earliest and most fundamental strategies by which humans explore the world. From infancy, we actively probe our physical environment and social landscape, seeking out experiences that reduce uncertainty and reveal structure. This talk will provide a gentle introduction to active learning as a core ingredient of learning behavior, tracing its classical motivation and exploring its modern transformation. In the first part, I will revisit active learning through the traditional lens of data efficiency, presenting a concrete method that illustrates its impact and core design principles. I will also outline a Bayesian perspective that reframes the relationship between passive learning and active querying. In the second part, I will turn to the challenges active learning faces in contemporary machine learning, particularly in large foundation models and continual learners. Together, I hope these perspectives will depict an updated view of active learning for the new era of AI.