A Review On Machine Learning Based Models For Identifying Potential Adversarial And Poisoning Attacks

Abstract

Machine Learning (ML) Has Revolutionized Data-driven Decision-making Across Sectors Such As Healthcare, Finance, Defense, And Cybersecurity. However, As Its Influence Grows, So Does Its Vulnerability To Adversarial And Poisoning Attacks. Adversaries Exploit The Weaknesses Of ML Models To Manipulate Outputs Or Degrade System Performance, Posing Significant Risks In Critical Applications. As A Result, Developing Machine Learning-based Models To Detect And Counter Such Attacks Has Become Essential For Building Secure And Trustworthy AI Systems. One Major Area Of Research That Has Emerged Is The Detection Of Poisoning For Android Systems Using Neural Networks Due To The Complexity Of Data Set Of Attacks. Several Approaches Have Been Used So Far For The Effective Classification Of Poisoning Attacks. The Paper Investigates The Different Contemporary Neural Network Based Approaches Used Thus Far In The Detection Of Poisoning Attacks. The Approaches Used And Their Findings Have Been Illustrated With Their Salient Points. Moreover, An Analysis In The Form Of Shortcoming In Previous Work Has Been Cited So As To Define A Problem Statement To Work Upon.

Keywords