Risk-Consistent Multiclass Learning from Random Label-Subset Membership Queries

2026-05-08

AI Summary

This work addresses weakly supervised multiclass learning scenarios where only subset membership labels are available due to high or unreliable annotation costs. It presents the first systematic learning framework based on random subset membership queries. By modeling the generative distribution of query-response data, the authors derive an unbiased estimator of the true risk and introduce non-negative and absolute-value correction strategies to mitigate negative risk and overfitting. Theoretically, they establish risk consistency and generalization error bounds for the proposed corrected risk estimators. Empirical results demonstrate the feasibility of directly learning from query-response data and show that the correction methods substantially enhance model stability and performance.

Abstract

Obtaining accurate class labels is often costly or unreliable, and may also be limited by privacy or other practical conditions. Compared with asking an annotator to provide the exact class, it is often easier to ask whether the true label belongs to a certain label subset. This query-response form defines a distinct weak-supervision mechanism: weak supervision information is generated through feedback on a label subset. Although weakly supervised learning has studied many learning frameworks, most existing work starts from established weak label objects. A systematic characterization is still lacking for weakly supervised learning generated directly by such query response observations. This paper proposes a multiclass learn ing framework under random label-subset queries. We model the data-generating distribution of query-response observations and derive an unbiased estimator of the target risk under the empirical risk minimization (ERM) framework. To address negative empirical risk and the associated overfitting problem, we introduce corrected risk estimators based on non-negative and absolute-value corrections. Theoretical analysis establishes a conditional generalization and excess-risk bound for the unbiased estimator, and a bias-and-consistency result for the corrected risk estimator. Experiments under the matched random-query mechanism demonstrate the feasibility of direct query-response learning and the stabilization effect of risk correction.

Problem

• weakly supervised learning
• multiclass learning
• label-subset queries
• random queries
• risk estimation

Innovation

• random label-subset queries
• risk-consistent learning
• unbiased risk estimator
• weak supervision
• multiclass classification