Evidence-driven ternary contrastive learning with hierarchical mamba fusion for robust multimodal intent recognition

Key Points

• Robust multimodal intent recognition is achieved using advanced contrastive learning techniques.
• Performance measures reveal a significant improvement of 30% over traditional methods in recognizing intents across audio and visual inputs.
• Analysis of multimodal datasets enhances the robustness of intent recognition processes, leading to more accurate predictions.
• The findings indicate that hierarchical fusion increases reliability, but further validation on diverse datasets is necessary.