Leveraging Self-Paced Curriculum Learning for Enhanced Modality Balance in Multimodal Conversational Emotion Recognition
Researchers have developed a new framework called Self-Paced Curriculum Learning (SPCL) to improve multimodal conversational emotion recognition. This approach addresses challenges like modality misalignment and imbalanced learning by dynamically guiding the training process from easier to more difficult instances. Experiments on the IEMOCAP and MELD datasets showed significant performance gains, with SPCL improving F1-scores by up to 6.6% on IEMOCAP and 10.4% on MELD, demonstrating its effectiveness as a plug-and-play module. AI
IMPACT Enhances multimodal AI's ability to understand nuanced human emotions in conversations.