NVIDIA Research: Auditing AI Models for Verified Deployment under Semantic Specifications

1 ยท NVIDIA Corporation ยท Sept. 1, 2021, 1:45 a.m.
โ—‹ Quality assurance and audits are necessary for deep learning models. Current AI models require large data sets for training or a designed reward function that must be optimized. Algorithmically, AI is prone to optimizing behaviors that were not intended by the human designer. To help combat this, the AuditAI framework was developed to help audit these problems, which increases safety and ethical use of deep learning models during deployment....