Advance : adversarial collaborative learning for detection and verification of artificially created examples
Loading...
Authors
Meeting name
Sponsors
Date
Journal Title
Format
Thesis
Subject
Abstract
Adversarial learning methods have gained significant popularity in generating deceptive yet convincingly authentic data. While these techniques have proven beneficial for advancing artificial intelligence, they also give rise to a pressing concern regarding the authenticity of information consumed by the general public, exemplified by the prevalence of deepfakes. Consequently, various approaches have been proposed to detect adversarial generated data, aiming to address this challenge. However, a significant proportion of these innovations rely on non-iterative feedforward designs, leading to the overarching concept of an arms race between machine learning and detection systems. In the context of my research, I introduce ADVANCE: Adversarial Collaborative Learning For Detection And Verification Of Artificially Created Examples, an adversarial recognition pipeline that embraces a continuous and collaborative learning paradigm to facilitate an arms race between a deepfake generator and deepfake detector. Specifically, given a generative system and a detection system, this research aims to discover whether it is possible to create a stable pipeline to look for an endpoint to this arms race. These experiments embody a computational analysis of the pipeline and ultimately, results indicate that it is stable enough to be used for further research in improvements of the detection of adversarial generated data.
Table of Contents
PubMed ID
Degree
M.S.