safetycritiquebearishConventional deep neural network architectures lack principled uncertainty quantification, which is a critical limitation for deployment in safety-critical domainsMachine Learning (Statistics)02 Aug 2026http://arxiv.org/abs/2607.28248v1