Get Started
Home
Topics
Search
Library
Research questionHow can we estimate deep neural network reliability under FP32 bit flips with fewer injections?Exhaustively injecting hardware faults into modern neural networks can require prohibitive numbers of experiments. Statistical injection lowers the cost but may still waste samples because bit flips have different effects depending on the corrupted floating-point value.
AI
Evaluation & Benchmarks
Machine Learning
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.TreeFI: Value-Aware Statistical Fault Injection for Deep Neural NetworksThe source presents TreeFI, a value-aware statistical injection methodology, and evaluates it on CNN and Transformer models using CIFAR-10 and ImageNet. Evidence includes comparison with statistical fault-injection baselines and an exhaustive activation-injection reference on ResNet8; reported budget reductions reach 72.1×, but the evidence is limited to the evaluated models, datasets, FP32 single-bit faults, and executed weight campaigns.research paper · Sep 4, 2026
Related questions
How can image-based deepfake detectors stay accurate on low-resolution inputs under tight compute budgets?How can high-resolution diffusion Transformers prune tokens without sacrificing image fidelity or predictable compute?How can CNN inference fit on bare-metal microcontrollers with severe memory, binary-size, and runtime limits?Can phase-transition counts during fine-tuning predict final test accuracy across architectures and distribution shifts?