Rich and Transparent Active Learning

A fundamental task of machine learning is prediction. Applications include detecting spam, recommending products, and diagnosing patients. Machine learning algorithms need to be trained on exemplars that are annotated by humans. The accuracy of the models often improves with the increased size of annotated data. Yet, annotating data takes time and effort. Active learning aims to minimize the annotation effort by enabling the algorithms to direct the human attention to the most informative exemplars. In traditional active learning approaches, algorithms are limited in the types of information they can acquire, and they often do not provide any rationale to the user as to why a particular exemplar is chosen for annotation. In this talk, I will describe our research on enriching the interaction between the algorithms and users for more effective training of predictive models.

Event Topic

Data Science