Assistive signals for deep neural network classifiers

Abstract

Deep Neural Networks are brittle in that small changes in the input can drastically affect their prediction outcome and confidence. Consequently, research in this area mainly focus on adversarial attacks and defenses. In this paper, we take an alternative stance and introduce the concept of Assistive Signals, which are perturbations optimized to improve a model’s confidence score regardless if it’s under attack or not. We analyze some interesting properties of these assistive perturbations and extend the idea to optimize them in the 3D space simulating different lighting conditions and viewing angles. Experimental evaluations show that the assistive signals generated by our optimization method increase the accuracy and confidence of deep models more than those generated by conventional methods that work in the 2D space. ‘Assistive Signals’ also illustrate bias of ML models towards certain patterns in real-life objects.

Collaborative Adversarial Training

“…Yet, due to the difference in aim, studies of hypocritical examples in [38] were mainly performed based on mis-classified benign examples according to their formal definition, while our work concerns local landscapes around all benign examples. Other related work include unadversarial examples [34] and assistive signals [31] that designed 3D textures to customize objects for better classifying them.…”