![[2507.05992] Exploring Partial Multi-Label Learning via Integrating Semantic Co-occurrence Knowledge](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2507.05992] Exploring Partial Multi-Label Learning via Integrating Semantic Co-occurrence Knowledge
Summary
This paper presents SCINet, a novel framework for partial multi-label learning that integrates semantic co-occurrence knowledge to improve label-instance relationship identification.
Why It Matters
Partial multi-label learning is crucial for effectively utilizing incompletely annotated data, which is common in real-world applications. The proposed SCINet framework enhances understanding of label relationships, potentially leading to better performance in various machine learning tasks, particularly in computer vision.
Key Takeaways
- SCINet introduces a bi-dominant prompter module to enhance semantic alignment between text and images.
- The framework models inter-label correlations and instance relationships to improve learning from ambiguous data.
- An intrinsic semantic augmentation strategy is proposed to boost model understanding through diverse image transformations.
- Extensive experiments show SCINet outperforms existing state-of-the-art methods on benchmark datasets.
- The research addresses a significant challenge in multi-label learning, making it relevant for practical applications.
Computer Science > Computer Vision and Pattern Recognition arXiv:2507.05992 (cs) [Submitted on 8 Jul 2025 (v1), last revised 23 Feb 2026 (this version, v2)] Title:Exploring Partial Multi-Label Learning via Integrating Semantic Co-occurrence Knowledge Authors:Xin Wu, Fei Teng, Yue Feng, Kaibo Shi, Zhuosheng Lin, Ji Zhang, James Wang View a PDF of the paper titled Exploring Partial Multi-Label Learning via Integrating Semantic Co-occurrence Knowledge, by Xin Wu and 6 other authors View PDF HTML (experimental) Abstract:Partial multi-label learning aims to extract knowledge from incompletely annotated data, which includes known correct labels, known incorrect labels, and unknown labels. The core challenge lies in accurately identifying the ambiguous relationships between labels and instances. In this paper, we emphasize that matching co-occurrence patterns between labels and instances is key to addressing this challenge. To this end, we propose Semantic Co-occurrence Insight Network (SCINet), a novel and effective framework for partial multi-label learning. Specifically, SCINet introduces a bi-dominant prompter module, which leverages an off-the-shelf multimodal model to capture text-image correlations and enhance semantic alignment. To reinforce instance-label interdependencies, we develop a cross-modality fusion module that jointly models inter-label correlations, inter-instance relationships, and co-occurrence patterns across instance-label assignments. Moreover, we propose...
