Hacking Neural Networks: A Short Introduction (2019-11-18T00:00:00.000000Z)

TL;DR

How deep learning in security works is given and the basic methods of exploitation are explored, but also the offensive capabilities deep learning enabled tools provide are looked at.

Abstract

A large chunk of research on the security issues of neural networks is focused on adversarial attacks. However, there exists a vast sea of simpler attacks one can perform both against and with neural networks. In this article, we give a quick introduction on how deep learning in security works and explore the basic methods of exploitation, but also look at the offensive capabilities deep learning enabled tools provide. All presented attacks, such as backdooring, GPU-based buffer overflows or automated bug hunting, are accompanied by short open-source exercises for anyone to try out.

Authors

Michael D Kissner

1 papers

TL;DR

Abstract

Authors

References71 items

NumPy

Iris Verification with Convolutional Neural Network and Unit-Circle Layer

Deep Leakage from Gradients

A Survey of Deep Learning Methods for Cyber Security

Backdooring Convolutional Neural Networks via Targeted Weight Perturbations

DeepLaser: Practical Fault Attack on Deep Neural Networks.

Fine-Pruning: Defending Against Backdooring Attacks on Deep Neural Networks

Adversarial Attacks and Defences Competition

A Survey of Techniques for Improving Security of GPUs

Invisible Mask: Practical Attacks on Face Recognition with Infrared

Microsoft Malware Classification Challenge

A Deep Learning Approach to Network Intrusion Detection

Adversarial Examples: Attacks and Defenses for Deep Learning

Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoning

Security Risks in Deep Learning Implementations

Security and Machine Learning

Robust Physical-World Attacks on Deep Learning Models

LOGAN: Membership Inference Attacks Against Generative Models

SecureML: A System for Scalable Privacy-Preserving Machine Learning

Deep Models Under the GAN: Information Leakage from Collaborative Deep Learning

Summoning Demons: The Pursuit of Exploitable Bugs in Machine Learning

Auror: defending against poisoning attacks in collaborative deep learning systems

Accessorize to a Crime: Real and Stealthy Attacks on State-of-the-Art Face Recognition

Adversarial examples in the physical world

CryptoNets: Applying Neural Networks to Encrypted Data with High Throughput and Accuracy

Practical Black-Box Attacks against Machine Learning

Asynchronous Methods for Deep Reinforcement Learning

Deep Residual Learning for Image Recognition

Malware classification with recurrent networks

CANN: An intrusion detection system based on combining cluster centers and nearest neighbors

Adam: A Method for Stochastic Optimization

Explaining and Harnessing Adversarial Examples

Going deeper with convolutions

Sequence to Sequence Learning with Neural Networks

ImageNet classification with deep convolutional neural networks

Deep Sparse Rectifier Neural Networks

Anomaly-based network intrusion detection: Techniques, systems and challenges

Digital selection and analogue amplification coexist in a cortex-inspired silicon circuit

Long Short-Term Memory

A neural network for speaker-independent isolated word recognition

Backpropagation Applied to Handwritten Zip Code Recognition

Language Models are Unsupervised Multitask Learners

Cuda c programming guide

Tatoeba

Cuda c best practices

Rapid7

Attack matrix

The state of machine learning frameworks in 2019, 2019

Kaggle

PRALab and Pluribus One s.r.l. Secml library

Machine learning’s privacy problem, 2019

Canadian Institute for Cybersecurity. Cybersecurity datasets

Trojaning Attack on Neural Networks

A short introduction to entropy, cross-entropy and kl-divergence

Program Synthesis from Natural Language Using Recurrent Neural Networks

Security and Privacy Weaknesses of Neural Networks

””3Blue1Brown”” Sanderson

Compilers Principles Techniques And Tools

Deep Learning

Chollet

An efficient intrusion detection system based on support vector machines and gradually feature removal method

Rmsprop: Divide the gradient by a running average of its recent magnitude

Hierarchical Data Format

http://www.cbsr.ia.ac.cn/china/Iris% 20Databases%20CH

Kdd cup 1999 data

Gradient-based learning applied to document recognition

Reinforcement Learning: An Introduction

Catastrophic Interference in Connectionist Networks: The Sequential Learning Problem

PyCuda is required for the GPU-based attack exercise. If no nVidia GPU is available on the machine

An editor is required to work with the code, preferably one that allows code highlighting for Python. Vim/Emacs will do. For reference, all exercises were prepared using Visual Studio Code

NLTK provides functionalities for natural language processing and is very helpful for some of the exercises

Field of Study

Journal Information

Name

Volume

Venue Information