Explore a 20-minute conference talk from PLDI 2023 that introduces a novel framework for incremental and complete verification of deep neural networks (DNNs). Learn about IVAN, a tool that achieves significant speedups in verifying challenging MNIST, CIFAR10, and ACAS-XU classifiers compared to state-of-the-art baselines. Discover how this approach improves efficiency when verifying updated DNNs, addressing the limitations of existing complete verifiers that require full re-verification. Gain insights into the innovative theory, data structures, and algorithms behind this framework, which aims to enhance the trustworthiness and robustness of DNNs in various applications.
Overview
Syllabus
[PLDI'23] Incremental Verification of Neural Networks
Taught by
ACM SIGPLAN
Related Courses
-
Mastering Neural Networks and Model Regularization
-
Foundations of Deep Learning and Neural Networks
-
Architecture-Preserving Provable Repair of Deep Neural Networks
-
Keras Deep Learning & Generative Adversarial Networks (GAN)
-
Advanced Generative Adversarial Networks (GANs)
-
Input-Relational Verification of Deep Neural Networks
