NNSmith: Generating Diverse and Valid Test Cases for Deep Learning Compilers

Keyphrases

• Compiler 100%
• Deep Learning Compiler 100%
• Deep Learning 50%
• Deep Neural Network 50%
• TensorRT 50%
• False Alarm 25%
• Corrupt 25%
• Floating-point 25%
• Testing Approaches 25%
• Resource Utilization 25%
• Compiler Transformation 25%
• Model Execution 25%
• Downstream Applications 25%
• Network Testing 25%
• Gradient Search 25%
• PyTorch 25%
• Differential Testing 25%
• Fuzz Testing 25%

Computer Science

• Deep Learning 100%
• Deep Neural Network 66%
• Neural Network Model 33%
• Floating Point 33%
• Resource Utilisation 33%
• Execution Model 33%
• Fuzz Testing 33%