Home // IARIA Congress 2023, The 2023 IARIA Annual Congress on Frontiers in Science, Technology, Services, and Applications // View article


The Triumvirate of Bespoke Diverse Hybridized Activation Functions, Adaptive Momentum, and Enhanced Entropic Wavelet Energy Spectrum Discernment for Higher Efficacy Detection of Artificial Intelligence-centric Attacks

Authors:
Steve Chan

Keywords: Artificial Intelligence Attack; Cyber; Smart Grid; Operational Technology; Adaptive Momentum; Second-Order Cone Programming Relaxations; Robust Convex Relaxations.

Abstract:
Artificial Intelligence-centric Attacks (AIA) involving False Data Injection and False Command Injection have become increasingly sophisticated and have also involved Metamorphic Malware (MM), which leverages numerous transformation techniques to avoid detection. Accordingly, the use of AI defender systems requires continuous learning so as to decrease the advantage held by the high cycles of adaptation of attackers. This paper explored Exponential Linear Unit (ELU) Mish as a hybridized activation function that could avoid the cessation of learning when a substantive portion of the Neural Network (NN) neurons output zero and weights are no longer updated. The involved Robust Convex Relaxation (RCR)-based Convolutional NN capitalized upon its architecture by utilizing a Nonlinear Conjugate Gradient and Nesterov’s Accelerated Gradient approach to Adaptive Momentum (AdaM) so as to mitigate against oscillation, facilitate convergence, and achieve a more optimal global minimum. This more resilient foundational architecture could then better support a more accurate and expedient Entropic [Wavelet Energy Spectrum] Discernment (ED). Central to this was the use of Second-Order Cone Programming Relaxations to address certain nonconvex subproblems, which were inadvertently spawned via the utilized RCR framework. The described triumvirate approach constitutes the beginnings of a potential mitigation pathway, which exhibited some promise during the preliminary experimentation.

Pages: 196 to 202

Copyright: Copyright (c) IARIA, 2023

Publication date: November 13, 2023

Published in: conference

ISBN: 978-1-68558-089-6

Location: Valencia, Spain

Dates: from November 13, 2023 to November 17, 2023