Fusion SHAP

Fusion SHAP leverages Shapley values, a game-theoretic concept, to explain the contributions of different features (e.g., from multiple data modalities like images and genomics) to a model's prediction. Current research focuses on improving SHAP's stability, efficiency, and applicability in diverse contexts, including long-tailed data, multimodal learning, and survival analysis, often employing techniques like kernel SHAP, sampling-based methods, and manifold projections to address limitations in high-dimensional spaces. This work is significant for enhancing the transparency and trustworthiness of complex machine learning models across various fields, from healthcare and finance to computer vision, by providing more reliable and interpretable explanations of model predictions.