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

Exploiting Negative Multi-Cluster Structure in Class-Wise Embeddings for Weakly Supervised Multi-Label Learning

NeurIPS 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Partial and multi-label learning

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.

Proposes exploiting the multi-cluster structure of negative samples in class-wise embeddings to improve weakly supervised multi-label learning.

Problem

Class-wise embedding methods for multi-label learning overlook the internal distribution of negative samples, which in weakly supervised settings limits discrimination.

Approach

Observes that negative instances in class-wise spaces form multiple semantically coherent sub-clusters and exploits this negative multi-cluster structure in the class-wise embedding learning.

Surprising findings

• Negative instances in class-wise spaces naturally organize into multiple semantically coherent sub-clusters rather than a single homogeneous group.

Experimental setup

Builds on: class-wise embeddings for multi-label learning

Closest papers by meaning

• Class-Distribution-Aware Pseudo-Labeling for Semi-Supervised Multi-Label Learning (NeurIPS 2023 · similarity 0.90)
• Self-Paced Unified Representation Learning for Hierarchical Multi-Label Classification (AAAI 2024 · similarity 0.88)
• Adaptive Integration of Partial Label Learning and Negative Learning for Enhanced Noisy Label Learning (AAAI 2024 · similarity 0.87)
• Rebalancing Multi-Label Class-Incremental Learning (AAAI 2025 · similarity 0.87)
• Weakly-Supervised Contrastive Learning for Imprecise Class Labels (ICML 2025 · similarity 0.87)
• Can Class-Priors Help Single-Positive Multi-Label Learning? (NeurIPS 2025 · similarity 0.87)