Neural network verification is an emerging approach to establish neural networks satisfy various specifications and requirements, such as formalized notions of safety, security, and trustworthiness. This tutorial will present neural network verification methods in the context of medical imaging analysis and aims to present this line of work being developed in the formal methods, security, and machine learning communities to the medical imaging analysis community. Interactive demonstrations will be done through the Neural Network Verification (NNV) software tool, and for example, allow for characterizations of robustness for neural networks used in medical imaging analysis.

Representative earlier tutorials on this approach for other domains are available here, which will be tailored to medical imaging analysis for this audience:
https://github.com/verivital/nnv/tree/master/code/nnv/examples/Tutorial

The tutorial will begin with an introductory lecture of about a half hour, followed by about an hour of interactive demonstrations of the NNV software tool. There are some technical pre-requisites for this workshop, particularly for the interactive part.