Evaluation of Neural Networks for Image Recognition Applications: Designing a 0-1 MILP Model of a CNN to create adversarials

TL;DR

This paper explores optimizing CNNs into capsule architectures, introduces a MILP model for generating adversarial examples, and evaluates various networks for image recognition tasks.

Contribution

It presents a MILP-based method to create adversarials and compares different CNN and capsule architectures for image recognition.

Findings

Capsule architectures outperform standard CNNs in certain tasks.

The MILP model effectively generates adversarial examples.

Evaluation results highlight strengths and weaknesses of each network type.

Abstract

Image Recognition is a central task in computer vision with applications ranging across search, robotics, self-driving cars and many others. There are three purposes of this document: 1. We follow up on (Fischetti & Jo, December, 2017) and show how standard convolutional neural network can be optimized to a more sophisticated capsule architecture. 2. We introduce a MILP model based on CNN to create adversarials. 3. We compare and evaluate each network for image recognition tasks.