Closing the BIG-LID: An Effective Local Intrinsic Dimensionality Defense for Nonlinear Regression Poisoning

Weerasinghe, Sandamal; Abraham, Tamas; Alpcan, Tansu; Erfani, Sarah M.; Leckie, Christopher; Rubinstein, Benjamin I. P.

doi:10.24963/ijcai.2021/437

Closing the BIG-LID: An Effective Local Intrinsic Dimensionality Defense for Nonlinear Regression Poisoning

Sandamal Weerasinghe, Tamas Abraham, Tansu Alpcan, Sarah M. Erfani, Christopher Leckie, Benjamin I. P. Rubinstein

Download BibTeX Cite

Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence. Main Track. Pages 3176-3184.

https://doi.org/10.24963/ijcai.2021/437

Keywords

Machine Learning: Adversarial Machine Learning
Data Mining: Anomaly/Outlier Detection
Multidisciplinary Topics and Applications: Security and Privacy

Abstract

Nonlinear regression, although widely used in engineering, financial and security applications for automated decision making, is known to be vulnerable to training data poisoning. Targeted poisoning attacks may cause learning algorithms to fit decision functions with poor predictive performance. This paper presents a new analysis of local intrinsic dimensionality (LID) of nonlinear regression under such poisoning attacks within a Stackelberg game, leading to a practical defense. After adapting a gradient-based attack on linear regression that significantly impairs prediction capabilities to nonlinear settings, we consider a multi-step unsupervised black-box defense. The first step identifies samples that have the greatest influence on the learner's validation error; we then use the theory of local intrinsic dimensionality, which reveals the degree of being an outlier of data samples, to iteratively identify poisoned samples via a generative probabilistic model, and suppress their influence on the prediction function. Empirical validation demonstrates superior performance compared to a range of recent defenses.

Sandamal Weerasinghe, Tamas Abraham, Tansu Alpcan, Sarah M. Erfani, Christopher Leckie, and Benjamin I. P. Rubinstein. 2021. Closing the BIG-LID: An Effective Local Intrinsic Dimensionality Defense for Nonlinear Regression Poisoning. In Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence, IJCAI-21. IJCAI, 3176–3184. https://doi.org/10.24963/ijcai.2021/437

@inproceedings{ijcai2021p437,
  title     = {Closing the BIG-LID: An Effective Local Intrinsic Dimensionality Defense for Nonlinear Regression Poisoning},
  author    = {Weerasinghe, Sandamal and Abraham, Tamas and Alpcan, Tansu and Erfani, Sarah M. and Leckie, Christopher and Rubinstein, Benjamin I. P.},
  booktitle = {Proceedings of the Thirtieth International Joint Conference on
               Artificial Intelligence, {IJCAI-21}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {Prof. Zhi-Hua Zhou},
  pages     = {3176--3184},
  year      = {2021},
  month     = {8},
  note      = {Main Track},
  doi       = {10.24963/ijcai.2021/437},
  url       = {https://doi.org/10.24963/ijcai.2021/437},
}

TY  - CPAPER
TI  - Closing the BIG-LID: An Effective Local Intrinsic Dimensionality Defense for Nonlinear Regression Poisoning
AU  - Weerasinghe, Sandamal
AU  - Abraham, Tamas
AU  - Alpcan, Tansu
AU  - Erfani, Sarah M.
AU  - Leckie, Christopher
AU  - Rubinstein, Benjamin I. P.
T2  - Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence, IJCAI-21
PB  - International Joint Conferences on Artificial Intelligence Organization
PY  - 2021
SP  - 3176
EP  - 3184
DO  - 10.24963/ijcai.2021/437
UR  - https://doi.org/10.24963/ijcai.2021/437
ER  -