Just How Toxic is Data Poisoning? A Unified Benchmark for Backdoor and Data Poisoning Attacks (2020-06-22T00:00:00.000000Z)

TL;DR

Unified benchmarks for data poisoning and backdoor attacks are developed in order to promote fair comparison in future work and to find that existing poisoning methods have been tested in contrived scenarios, and they fail in realistic settings.

Abstract

Data poisoning and backdoor attacks manipulate training data in order to cause models to fail during inference. A recent survey of industry practitioners found that data poisoning is the number one concern among threats ranging from model stealing to adversarial attacks. However, we find that the impressive performance evaluations from data poisoning attacks are, in large part, artifacts of inconsistent experimental design. Moreover, we find that existing poisoning methods have been tested in contrived scenarios, and they fail in realistic settings. In order to promote fair comparison in future work, we develop unified benchmarks for data poisoning and backdoor attacks.

Authors

John P. Dickerson

6 papers

Micah Goldblum

6 papers

Avi Schwarzschild

7 papers

TL;DR

Abstract

Authors

References51 items

VENOMAVE: Clean-Label Poisoning Against Speech Recognition

Don't Trigger Me! A Triggerless Backdoor Attack Against Deep Neural Networks

Witches' Brew: Industrial Scale Data Poisoning via Gradient Matching

Min-Max Optimization without Gradients: Convergence and Applications to Black-Box Evasion and Poisoning Attacks

Benchmarking Adversarial Robustness on Image Classification

Bullseye Polytope: A Scalable Clean-Label Poisoning Attack with Improved Transferability

Poisoning Attacks on Algorithmic Fairness

MetaPoison: Practical General-purpose Clean-label Data Poisoning

Dynamic Backdoor Attacks Against Machine Learning Models

Influence Function based Data Poisoning Attacks to Top-N Recommender Systems

Adversarial Machine Learning-Industry Perspectives

Targeted Poisoning Attacks on Social Recommender Systems

Latent Backdoor Attacks on Deep Neural Networks

Deep Learning Poison Data Attack Detection

Justifying Recommendations using Distantly-Labeled Reviews and Fine-Grained Aspects

Hidden Trigger Backdoor Attacks

Deep k-NN Defense Against Clean-Label Data Poisoning Attacks

Min-Max Optimization without Gradients: Convergence and Applications to Adversarial ML

A Backdoor Attack Against LSTM-Based Text Classification Systems

Transferable Clean-Label Poisoning Attacks on Deep Neural Nets

Stronger data poisoning attacks break data sanitization defenses

The Open Images Dataset V4

Clean-Label Backdoor Attacks

Poisoning Attacks to Graph-Based Recommender Systems

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

Poison Frogs! Targeted Clean-Label Poisoning Attacks on Neural Networks

MobileNetV2: Inverted Residuals and Linear Bottlenecks

Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoning

MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks on Corrupted Labels

Towards Poisoning of Deep Learning Algorithms with Back-gradient Optimization

BadNets: Identifying Vulnerabilities in the Machine Learning Model Supply Chain

Towards Deep Learning Models Resistant to Adversarial Attacks

The Marginal Value of Adaptive Gradient Methods in Machine Learning

Understanding Black-box Predictions via Influence Functions

Data Poisoning Attacks on Factorization-Based Collaborative Filtering

Deep Residual Learning for Image Recognition

Support vector machines under adversarial label contamination

Very Deep Convolutional Networks for Large-Scale Image Recognition

ImageNet classification with deep convolutional neural networks

Poisoning Attacks against Support Vector Machines

Et al

Amazon Review Data

Trojaning Attack on Neural Networks

Just How Toxic is Data Poisoning? A Uniﬁed Benchmark for Backdoor and Data 29th International Coference on International Conference on Machine Learning , ICML’12, pp. 1467–1474, 2012

Tiny ImageNet Visual Recognition Challenge

Tiny ImageNet Visual Recognition Challenge

I and J

Learning Multiple Layers of Features from Tiny Images

Influence of cultivation temperature on the ligninolytic activity of selected fungal strains

AND T

Just How Toxic is Data Poisoning? A Uniﬁed Benchmark for Backdoor and Data Poisoning Attacks

Field of Study

Journal Information

Name

Volume

Venue Information

Name

Type

URL

Alternate Names