Generative Models for Security: Attacks, Defenses, and Opportunities

View on arXiv ← Back to list

Authors: Luke A. Bauer, Vincent Bindschaedler

Published: 2021-07-21 15:16:10+00:00

AI Summary

This research paper surveys recent research at the intersection of security and privacy with generative models. It focuses on the use of generative models in adversarial machine learning, attacks, and defenses across various cybersecurity contexts like intrusion detection and deepfake creation, also highlighting opportunities in data synthesis and fairness.

Abstract

Generative models learn the distribution of data from a sample dataset and can then generate new data instances. Recent advances in deep learning has brought forth improvements in generative model architectures, and some state-of-the-art models can (in some cases) produce outputs realistic enough to fool humans. We survey recent research at the intersection of security and privacy and generative models. In particular, we discuss the use of generative models in adversarial machine learning, in helping automate or enhance existing attacks, and as building blocks for defenses in contexts such as intrusion detection, biometrics spoofing, and malware obfuscation. We also describe the use of generative models in diverse applications such as fairness in machine learning, privacy-preserving data synthesis, and steganography. Finally, we discuss new threats due to generative models: the creation of synthetic media such as deepfakes that can be used for disinformation.

Key findings

Generative models significantly impact security and privacy, enhancing both attacks (e.g., creating realistic deepfakes and adversarial examples) and defenses (e.g., improving anomaly detection and creating synthetic datasets). The survey reveals a growing body of research in this area, highlighting open challenges in evaluating generative models and detecting synthetic media.

Approach

The paper conducts a comprehensive survey of existing literature, categorizing the use of generative models (GANs, VAEs, autoregressive models) in security and privacy applications. It analyzes attacks and defenses enabled by these models, examining their effectiveness and limitations across different modalities (audio, video, text).

Datasets

UNKNOWN

Model(s)

Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), Autoregressive models (PixelRNN, PixelCNN, WaveNet), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTMs), Transformers

Author countries

USA

← Previous