可微信念函数层：用Yager组合规则构建深度证据神经网络
原标题：Neural Networks with Differentiable Belief-Function Layers
AI 摘要
该论文提出一种可微的信念函数层，将Yager 1987年的组合规则用于深度证据神经网络（MP-ENN），使网络中每一层质量函数均为合法的Dempster–Shafer质量函数。作者指出Dempster规则中的归一化项在证据冲突时会导致梯度发散，而Yager规则是双线性的、梯度有界，因此更适合反向传播。实验显示该网络仅用硬标签训练，却能在视觉（CIFAR-10H）和语言两种模态上对人类分歧进行排序，其输出中的无知集质量可作为“我不知道”信号，并在分布外检测上具有竞争力。
正文节选
Neural Networks with Differentiable Belief-Function LayersThanks: Code: https://github.com/sulcantonin/enn-public Abstract Yager’s 1987 rule of combination was proposed on epistemic grounds: under high conflict, mass should be redistributed to ignorance rather than renormalised onto the class simplex. We observe that backpropagation needs the same rule: Dempster’s normaliser , where is the conflict mass, puts a term in the gradient that diverges exactly when evidence sources disagree, while Yage