Memory-guided Prototypical Co-occurrence Learning for Mixed Emotion Recognition
New AI framework uses memory banks to detect complex, overlapping human emotions from physiological signals.
A research team from multiple institutions has introduced a novel AI framework called Memory-guided Prototypical Co-occurrence Learning (MPCL) that significantly advances the field of affective computing. Published on arXiv, the work addresses a critical limitation in current emotion recognition systems, which typically predict singular, basic emotions in controlled lab settings. Real human emotional experience is far more complex, often involving overlapping affective states like feeling both happy and sad simultaneously. The MPCL framework is designed specifically for this 'mixed emotion recognition' problem, treating it as an emotion distribution learning task rather than simple classification.
The technical innovation lies in MPCL's hierarchical architecture that mimics aspects of human cognitive memory. It first fuses multi-modal physiological and behavioral signals through a multi-scale associative memory mechanism. Then, it constructs emotion-specific prototype memory banks to capture cross-modal semantic relationships, using a technique called prototype relation distillation to ensure alignment in the latent space. A key component is a memory retrieval strategy that extracts semantic-level co-occurrence associations across emotion categories. This bottom-up process allows the model to learn representations that capture the structured correlations and valence consistency inherent in coexisting emotions. Comprehensive experiments demonstrate MPCL's superior performance over existing methods, paving the way for more nuanced and realistic human-AI interaction systems.
- The MPCL framework treats emotion recognition as a distribution learning problem, moving beyond single-label classification.
- It uses emotion-specific prototype memory banks and a retrieval strategy to model the co-occurrence patterns of mixed emotions.
- Outperforms state-of-the-art methods on two public datasets, showing both quantitative and qualitative improvements.
Why It Matters
Enables AI to understand complex, real-world human emotions, improving mental health apps, empathetic chatbots, and human-computer interaction.