Brief

Researchers have introduced Neural Bayesian Sequential Routing (NBSR), a novel framework that models neural inference as active evidence accumulation within a hierarchical Directed Acyclic Graph (DAG). This approach allows neural networks to mimic human decision-making by acquiring evidence, updating belief states, and determining when to stop computation. NBSR offers uncertainty quantification, early exiting, and cost-aware evidence acquisition, demonstrating competitive performance across various tasks including visual categorization, language modeling, and experimental design. AI