Spartan Networks: Self-feature-squeezing neural networks for increased robustness in adversarial settings

Abstract

Deep learning models are vulnerable to adversarial examples which are input samples modified in order to maximize the error on the system. We introduce Spartan Networks, resistant deep neural networks that do not require input preprocessing nor adversarial training. These networks have an adversarial layer designed to discard some information of the network, thus forcing the system to focus on relevant input. This is done using a new activation function to discard data. The added layer trains the neural network to filter-out usually-irrelevant parts of its input. Our performance evaluation shows that Spartan Networks have a slightly lower precision but report a higher robustness under attack when compared to unprotected models. Results of this study of Adversarial AI as a new attack vector are based on tests conducted on the MNIST dataset. KEYWORDSArtificial Intelligence, Cybersecurity, Adversarial AI Potentially any deep learning model can be vulnerable to this kind of attack, which is hard to detect, prevent, and whose impact will only grow in the upcoming years.

Towards a Robust Deep Neural Network in Texts: A Survey

“…Both of them are long and arduous tasks with the common efforts of many people. At the moment, the defender can draw on methods from image area to text for improving the robustness of DNNs, e.g., adversarial training [162], adding extra layer [175], optimizing cross-entropy function [176], [177], or weakening the transferability of adversarial examples.…”

Türki̇ye Süper Li̇g Sezon Sonu Takim Siralamasinin Geli̇şti̇ri̇len Yapay Si̇ni̇r Ağlari Modeli̇ İle Tahmi̇n Edi̇lmesi̇

“…Bir nöron kendisine gelen girdi değerini değerlendirerek eşik değerine bağlı olarak ateşleme yapılıp yapılmamasına karar verebilen bir mekanizmaya sahiptir (Şekil 1). Nöronlar, aralarında kurulan iletişim sayesinde karar verebilirler [17].…”