Safety Verification via Deep Learning: A conceptual framework

---

Authors:

Prof. Sanjoy Baruah

Abstract:

"Computational resources are scarce upon many safety-critical edge devices. Such devices often deal with highly dynamic workloads that evolve during execution, requiring safety properties to be re-verified repeatedly at runtime. Since verifying such properties can be computationally highly intractable, it is reasonable to consider training deep-learning (DL) classifiers to classify system specifications according to whether they satisfy relevant safety requirements. A framework will be proposed for doing so in a manner that does not compromise system safety despite the inherent possibility of occasional errors in DL-based classification."