---
title: "DEEPSPLIT: An Efficient Splitting Method for Neural Network Verification via Indirect Effect Analysis"
authors:
  - "Patrick Henriksen"
  - "Alessio Lomuscio"
doi: "10.24963/ijcai.2021/351"
url: "https://ijcai.org/proceedings/2021/351"
pdf: "https://ijcai.org/proceedings/2021/0351.pdf"
venue: "Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence"
year: 2021
pages: "2549-2555"
track: "Main Track"
keywords:
  - "Machine Learning: Deep Learning"
  - "Multidisciplinary Topics and Applications: Validation and Verification"
---

# DEEPSPLIT: An Efficient Splitting Method for Neural Network Verification via Indirect Effect Analysis

Patrick Henriksen, Alessio Lomuscio

*Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence*, pages 2549-2555. DOI: [10.24963/ijcai.2021/351](https://doi.org/10.24963/ijcai.2021/351)

## Abstract

We propose a novel, complete algorithm for the verification and analysis of feed-forward, ReLU-based neural networks. The algorithm, based on symbolic interval propagation, introduces a new method for determining split-nodes which evaluates the indirect effect that splitting has on the relaxations of successor nodes. We combine this with a new efficient linear-programming encoding of the splitting constraints to further improve the algorithm’s performance. The resulting implementation, DeepSplit, achieved speedups of 1–2 orders of magnitude and 21-34% fewer timeouts when compared to the current SoA toolkits.