Search papers, labs, and topics across Lattice.
SimCert is introduced as a probabilistic certification framework to verify the behavioral similarity between original and compressed DNNs, addressing the challenge of deploying DNNs on resource-constrained systems while maintaining safety. It uses a dual-network symbolic propagation method that supports both quantization and pruning, combined with a variance-aware bounding technique based on Bernstein's inequality to tighten safety certificates. Experiments on ACAS Xu and computer vision benchmarks show SimCert outperforms existing verification methods, providing quantitative safety guarantees with adjustable confidence levels.
Get probabilistic safety guarantees for compressed neural networks with SimCert, a new framework that beats state-of-the-art verification methods.
Deploying Deep Neural Networks (DNNs) on resource-constrained embedded systems requires aggressive model compression techniques like quantization and pruning. However, ensuring that the compressed model preserves the behavioral fidelity of the original design is a critical challenge in the safety-critical system design flow. Existing verification methods often lack scalability or fail to handle the architectural heterogeneity introduced by pruning. In this work, we propose SimCert, a probabilistic certification framework for verifying the behavioral similarity of compressed neural networks. Unlike worst-case analysis, SimCert provides quantitative safety guarantees with adjustable confidence levels. Our framework features: (1) A dual-network symbolic propagation method supporting both quantization and pruning; (2) A variance-aware bounding technique using Bernstein's inequality to tighten safety certificates; and (3) An automated verification toolchain. Experimental results on ACAS Xu and computer vision benchmarks demonstrate that SimCert outperforms state-of-the-art baselines.
