Discriminative Feature Feedback with General Teacher Classes

2025-10-08

AI Summary

This paper establishes the theoretical foundations of the Discriminative Feature Feedback (DFF) interactive learning protocol. We address both realizable and agnostic settings, introducing novel structural dimensions to characterize DFF’s expressive power. For the realizable case, we derive the first complete characterization of the optimal error bound; for the agnostic case, we provide a tight upper bound on the error rate and prove its optimality in general. Crucially, we demonstrate that classical complexity measures—such as VC dimension or Littlestone dimension—are insufficient to capture either the optimal achievable error or the feasibility of no-regret learning under DFF. Methodologically, our analysis integrates discriminative feature modeling, structured dimension definitions, optimal bound derivation, and constructive upper-bound techniques. Our results yield the first unified, rigorous theoretical framework for feedback-rich learning paradigms, bridging gaps between interactive, supervised, and online learning theories.

Technology Category

• Machine Learning: Learning Theory
• Search and Optimization: Learning to Search
• Game Theory and Economic Paradigms: Adversarial Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems

Abstract

We study the theoretical properties of the interactive learning protocol Discriminative Feature Feedback (DFF) (Dasgupta et al., 2018). The DFF learning protocol uses feedback in the form of discriminative feature explanations. We provide the first systematic study of DFF in a general framework that is comparable to that of classical protocols such as supervised learning and online learning. We study the optimal mistake bound of DFF in the realizable and the non-realizable settings, and obtain novel structural results, as well as insights into the differences between Online Learning and settings with richer feedback such as DFF. We characterize the mistake bound in the realizable setting using a new notion of dimension. In the non-realizable setting, we provide a mistake upper bound and show that it cannot be improved in general. Our results show that unlike Online Learning, in DFF the realizable dimension is insufficient to characterize the optimal non-realizable mistake bound or the existence of no-regret algorithms.

Problem

• Analyzing theoretical properties of Discriminative Feature Feedback protocol
• Characterizing mistake bounds in realizable and non-realizable settings
• Comparing DFF with classical learning protocols like online learning

Innovation

• Discriminative Feature Feedback protocol for interactive learning
• Mistake bound analysis in realizable and non-realizable settings
• Novel dimension concept characterizes DFF learning performance