New CONFER framework improves multimodal emotion recognition by negotiating conflicting data

Researchers have developed CONFER, a novel framework designed to improve multimodal emotion recognition by addressing the unreliability of self-reported labels and conflicts between different data modalities. This graph-based approach negotiates evidence among modality experts, estimating their predictive beliefs, uncertainties, and runtime reliability. CONFER categorizes samples into Consensus, Dissent, and Ambiguity regimes for weak-label calibration, demonstrating competitive performance with accuracies of 0.873 on AMIGOS-V and 0.854 on MAHNOB-V under strict leave-one-subject-out protocols. The framework effectively utilizes cross-modal conflict information for coordination and supervision-reliability estimation, showing increased robustness to corrupted weak labels. AI

IMPACT Enhances multimodal emotion recognition accuracy by effectively handling conflicting data and unreliable labels.

RANK_REASON The cluster contains a research paper detailing a new framework for multimodal emotion recognition. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• AMIGOS
• arXiv
• DEAP
• Hugging Face
• MAHNOB-HCI
• MAHNOB-V

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →