Event-Conditioned Credibility for Decision-Level Evidence Fusion
A novel technique for fusing heterogeneous evidence sources with varying precision and possible anomalies has been introduced. This method incorporates event-conditioned credibility to assess the relative trustworthiness of evidence across different candidate-event hypotheses. It features a decision-focused joint optimization model that integrates credibility assessment, evidence fusion, and event determination via candidate-event probabilities. Represented as a continuous self-mapping on the probability simplex, the model employs a direct fixed-point iteration as its primary fast solver, alongside a Kuhn simplicial search for approximating fixed points dependent on the mesh. The study tackles the issue where evidence that diverges from the majority could either be crucial for the right decision or misleading, a nuance often overlooked by current credible evidence fusion techniques.
Key facts
- Paper arXiv:2504.04128v2 proposes event-conditioned credibility for evidence fusion.
- Method addresses heterogeneous sources with unequal precision and potential anomalies.
- Decision-oriented joint optimization model couples credibility, fusion, and decision.
- Model expressed as continuous self-mapping on probability simplex.
- Default solver uses direct fixed-point iteration.
- Kuhn simplicial search provides mesh-dependent approximate fixed points.
- Existing methods may underestimate critical evidence deviating from majority.
- Approach distinguishes critical evidence from anomalous evidence.
Entities
—