Research map / Label-efficient and robust learning / Recognition and long-tail

Consistent Multi-Class Classification from Multiple Unlabeled Datasets

ICLR 2024 · Label-efficient and robust learning › Recognition and long-tail · approach: Learning from unlabeled and positive-unlabeled data

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Studies multi-class classification learned from multiple unlabeled datasets with different class priors, proposing a consistent learning approach for this weakly supervised setting.

Problem

Learning an accurate multi-class classifier from multiple unlabeled datasets, each with known class priors, where no labeled examples are available.

Approach

Uses multiple unlabeled datasets with different class priors and derives a consistent risk estimator for multi-class classification, extending prior weakly supervised (unlabeled-data with class priors) methods.

Experimental setup

Builds on: learning from unlabeled data with class priors

Closest papers by meaning

• Multi-class classification without multi-class labels (ICLR 2019 · similarity 0.88)
• Classification from Pairwise Similarity and Unlabeled Data (ICML 2018 · similarity 0.88)
• A General Formulation for Safely Exploiting Weakly Supervised Data (AAAI 2018 · similarity 0.88)
• AUC Optimization from Multiple Unlabeled Datasets (AAAI 2024 · similarity 0.88)
• Learning from Label Proportions: A Mutual Contamination Framework (NeurIPS 2020 · similarity 0.87)
• Learning with Complementary Labels Revisited: The Selected-Completely-at-Random Setting Is More Practical (ICML 2024 · similarity 0.86)