Leveraging Cadence's Incisive Enterprise Simulator for Neural Network Verification

EasyChair Preprint 15065

Abstract

As neural networks become increasingly integral to modern technology, ensuring their reliability and safety has emerged as a critical challenge. This paper explores the application of Cadence's Incisive Enterprise Simulator as a robust solution for neural network verification. The simulator offers advanced features such as high-performance mixed-signal simulation and support for formal verification techniques, making it well-suited for validating the complex architectures of neural networks. We discuss the modeling of neural networks as hardware components, the verification of functional correctness, and the utilization of formal methods to ensure critical properties like safety and robustness. Additionally, we highlight successful case studies that illustrate the simulator's effectiveness in verifying deep learning accelerators and safety-critical applications in autonomous systems. The paper concludes with insights into future enhancements and the potential for tighter integration with machine learning frameworks, emphasizing the growing need for sophisticated verification tools in the rapidly evolving field of artificial intelligence.

Keyphrases: Cadence Incisive Enterprise Simulator, Neural Network Verification, debugging techniques, formal verification

@booklet{EasyChair:15065,
  author    = {Harold Jonathan and Edwin Frank},
  title     = {Leveraging Cadence's Incisive Enterprise Simulator for Neural Network Verification},
  howpublished = {EasyChair Preprint 15065},
  year      = {EasyChair, 2024}}