Improving Machine Learning Beyond the Algorithm

User interaction data is at the heart of interactive machine learning systems (IMLSs), such as voice-activated digital assistants, e-commerce destinations, news content hubs, and movie streaming portals. In my talk, I will show how we can improve machine learning in such systems through a principled treatment of biases in interaction data via causal inference and counterfactual learning, and through interface interventions that increase the quality and quantity of interaction data from users. All these efforts are part of my larger vision that improving machine learning accuracy in IMLSs is not only a question of improving machine learning algorithms, but that there are also numerous other crucial questions, such as how interfaces affect interaction data quality and quantity.

Date:
Speakers:
Tobias Schnabel
Affiliation:
Cornell University

Watch Next