Where and when: Thursday, Aug 26 at 2-3pm in 303S-561
Speaker: Zac Pullar-Strecker, B.Sc. Student and Research Assistant (for Joerg Wicker)
Abstract: Training modern ML models frequently requires large datasets which are expensive and time-consuming to collect. Active Learning addresses this by providing a method to train models with smaller amounts of labeled data. However, the lack of data also makes it difficult for practitioners to determine when the model is ‘good enough’. We will introduce the structure & design of stopping criteria, cover a few of the most effective prior examples, describe our own method, and discuss strategies for evaluating multi-objective problems.