Towards Scalable Complete Verification of Relu Neural Networks via Dependency-based Branching

Authors: Panagiotis Kouvaros, Alessio Lomuscio

IJCAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the method on all of the Re LU-based fully connected networks from the first competition for neural network verification. The experimental results obtained show 145% performance gains over the present state-of-the-art in complete verification.
Researcher Affiliation Academia Panagiotis Kouvaros and Alessio Lomuscio Department of Computing, Imperial College London, UK {p.kouvaros, a.lomuscio}@imperial.ac.uk
Pseudocode Yes Algorithm 1 The verification procedure.
Open Source Code Yes Venus2 is available at https://github.com/vas-group-imperial/ venus2.
Open Datasets Yes ACASXU [Julian et al., 2016] is a collection of 45 Re LU FFNNs which were developed as part of an airborne collision avoidance system to advise horizontal steering decisions for unmanned aircraft. MNIST [Le Cun et al., 1998] is a dataset comprising images of hand-written digits 0-9, each formatted as 28x28x1-pixel grayscale image.
Dataset Splits No The paper mentions using pre-trained models (MNIST2, MNIST4, MNIST6) and verifying them against robustness properties for '25 correctly classified images' but does not specify the train/validation/test splits used for the training of these networks or for the verification experiments themselves beyond the '25 images'.
Hardware Specification Yes All the experiments were carried out on an Intel Core i77700K (4 cores) equipped with 16GB RAM, running Linux kernel 4.15; the ablation experiment on the branching depth (see below) was carried out on an Intel Core i9-9900X (20 cores) equipped with 125GB RAM, running Linux kernel 5.3.
Software Dependencies Yes All MILP-based tools (Venus2, Eran, MIPVerify, Venus) rely on Gurobi 9.2 for the MILP backend.
Experiment Setup Yes Venus2 was run with the branching depth set to 4 for MNIST2 and MNIST4 and to 7 for ACASXU and MNIST6. The B&B node threshold was set to 500 for ACASXU and MNIST6, to 5000 for MNIST4 and to 10000 for MNIST2.