Skip to main content
Loading Events

« All Events

  • This event has passed.

ECE Rising Stars Seminar: Steve Mussmann

March 3, 2023 @ 12:00 PM 1:00 PM

Data-efficient machine learning through active learning and data pruning

Abstract: Enormous datasets, with up to billions of images, have fueled recent progress in machine learning. At the same time, training over such large datasets is very computationally expensive. Furthermore, the best task-specific performance is achieved from using expensively labeled domain data. In this talk, I will present my research on decreasing machine learning data costs by selecting the most important data points. I will discuss insights for when and how the most common active learning algorithm, uncertainty sampling, yields strong performance including a case study where 14x less labeled data is required. I will then present theoretical and empirical results for machine teaching applied to data pruning, a technique to save training time by only training on the most important data points.

Steve Mussmann

Bio: Steve Mussmann is an IFDS Postdoctoral Fellow in the Paul G. Allen School of Computer Science & Engineering at the University of Washington working with Kevin Jamieson and Ludwig Schmidt on mitigating costs in machine learning by selecting the most important data points for labeling and training. He received a Ph.D. in 2021 from Stanford University in computer science advised by Percy Liang and a B.S. in 2015 from Purdue University in math, statistics, and computer science.

In addition to the in-person seminar, this presentation will be available live via Zoom with Meeting ID: 988 5990 0508

1415 Engineering Drive
Madison, 53711
+ Google Map