Learning from preferences and labels

Abstract: As machine learning is used for solving complex problems, eliciting meaningful labels and rewards for supervision is becomes challenging. Preferences in the form of pairwise comparisons have emerged as an alternate feedback mechanism that are often easier to elicit and more accurate. This talk will outline our efforts in understanding the fundamental limits of learning when an algorithm is given access to both preferences and labels. We will discuss and contrast the value of preferences in several settings including classification, regression, bandits, optimization and reinforcement learning, as time permits, along with some open problems.