ConfSleepNet: Conflict-Aware AI for Sleep Stage Classification
A new AI framework, ConfSleepNet, addresses the challenge of unreliable sleep stage classification when multi-modal data (e.g., EEG, EOG) are misaligned. Proposed in arXiv:2605.17021, the conflict-aware evidential framework dynamically resolves inter-view conflicts through two phases: multi-view evidence extraction and conflict-aware aggregation. The first phase learns category-related evidence from different modalities using hybrid category structures tailored to each modality's characteristics. The second phase constructs view-specific opinions—including prediction results and uncertainty—to improve reliability. The framework aims to enhance real-world applicability where perfect alignment of modalities is unattainable.
Key facts
- ConfSleepNet is a conflict-aware evidential framework for sleep stage classification.
- It uses multi-view learning with multi-modal data.
- The framework addresses misalignment between modalities in real-world scenarios.
- It consists of multi-view evidence extraction and conflict-aware aggregation.
- Hybrid category structures are proposed for different modalities.
- View-specific opinions include prediction results and uncertainty.
- The paper is published on arXiv with ID 2605.17021.
- The method aims to improve reliability of sleep staging results.
Entities
Institutions
- arXiv