August 5, 2026 Training with Confidence: Catching Silent Errors in Deep Learning Training with Automated Proactive Checks TRAINCHECK — 通过自动推断训练不变量并主动检查,在训练过程中提前发现静默错误(OSDI 2025, University of Michigan) OSDI-2025 DL-Training Silent-Errors Invariant-Inference 训练可靠性