Abstract
背景:The correctness of DL systems is crucial for trust in DL applications
本文: NeuRI
BaseTool: FreeFuzz
Github: https://github.com/ise-uiuc/neuri-artifact
Task: Fuzzing DL Libraris and explore more types of operators
Step:
- collecting valid and invalid API traces from various sources, 就是freefuzz
- apply inductive program synthesis over traces to infer the constraints
- hybrid fuzzing
实验:
对象:PyTorch, TensorFlow
Competitor: NNSmith
效果:
- +51%, 15% branch coverage
- detected 87 bugs, 64 confirmed, 8 high-priority bugs
- Diversifying Generation CDeepFuzz Inductive Inferencediversifying generation cdeepfuzz inductive understanding pre-trained generation cdeepfuzz coverage-guided generation cdeepfuzz coverage inductive diversifying inference mlcommons-inference inference computer learning vision inference paddle classification exploiting questions inference