Research map / Label-efficient and robust learning / Noisy labels

Not All Instances Should Contribute Equally: Sparse Support Grounding via Unbalanced Transport for Heterogeneous MIL

NeurIPS 2026 · Label-efficient and robust learning › Noisy labels · approach: Loss correction and noise robustness theory

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 Sparse Support Grounding, a multiple instance learning framework that learns per-instance influence via unbalanced optimal transport so that only a few supporting instances contribute to bag-level predictions.

Problem

Multiple instance learning from bag-level labels where only a few hidden instances support the bag label, and heterogeneous bags contain rare, diverse useful instances amid many weak or noisy ones.

Approach

Instead of letting any instance contribute by being matchable to a prototype, the method learns instance influence directly, using unbalanced transport to ground sparse support so that instances contribute only when they genuinely support the bag label.

Closest papers by meaning

• Sm: enhanced localization in Multiple Instance Learning for medical imaging classification (NeurIPS 2024 · similarity 0.86)
• Do Multiple Instance Learning Models Transfer? (ICML 2025 · similarity 0.84)
• WISH: Weakly Supervised Instance Segmentation using Heterogeneous Labels (CVPR 2025 · similarity 0.84)
• xMIL: Insightful Explanations for Multiple Instance Learning in Histopathology (NeurIPS 2024 · similarity 0.84)
• Coherent Hierarchical Multi-Label Learning to Defer for Medical Imaging (NeurIPS 2026 · similarity 0.84)
• Learning from Label Proportions: Bootstrapping Supervised Learners via Belief Propagation (ICLR 2024 · similarity 0.84)