Abstract
Training machine learning models usually involves iteratively optimizing the model using a uniformly randomly selected subset of samples. In contrast, effective human learning often relies on a carefully designed curriculum of learning materials adapted based on the learner’s progress (e.g., starting with learning basic arithmetic operations and gradually moving on to calculus). We can apply this intuition to machine learning by creating curricula with carefully arranged samples to achieve both better learning efficiency and performance. Lastly, I will share my experiences about how adapting my curriculum as an undergraduate/graduate student can benefit both learning and research.
Biography
Shengjie Wang is an assistant professor in computer science. His research focuses on machine learning in general, including deep learning, optimization, and AI for science. Shengjie Wang obtained his Ph.D. degree from the CSE department at the University of Washington under the supervision of Jeffrey Bilmes. Before joining NYU-SH, Shengjie Wang worked at Bytedance Seattle as a research scientist, focusing on recommendation systems and AI for science.