2021

Baseline Pruning-Based Approach to Trojan Detection in Neural Networks

Bajcsy, Peter, Majurski, Michael

Understand

This paper addresses the problem of detecting trojans in neural networks (NNs) by analyzing systematically pruned NN models.

Our pruning-based approach consists of three main steps.
First, detect any deviations from the reference look-up tables of model file sizes and model graphs.
Next, measure the accuracy of a set of systematically pruned NN models following multiple pruning schemas.

Built on

B. Hassibi, D. G. Stork, Second Order Derivatives for Network Pruning: Optimal Brain Surgeon, in: Advances in Neural Information Processing Systems 5 (NIPS 1992), Neural Information Processing Systems Foundation, Inc., 1992, pp. 164–172
1992
Earlier work this paper cites.
arXiv:2003.03033
Original
D. Blalock, J. J. G. Ortiz, J. Frankle, J. Guttag, What is the state of neural network pruning?, arXiv (2020) · 2003
Earlier work this paper cites.
arXiv:2008.00123
Original
N. B. Erichson, D. Taylor, Q. Wu, M. W. Mahoney, Noise-response analysis for rapid detection of backdoors in deep neural networks, arXiv (2020) · 2008
Earlier work this paper cites.
arXiv:1506.02626
Original
S. Han, J. Pool, J. Tran, W. J. Dally, Learning both Weights and Connections for Efficient Neural Networks (2015) · 2015
Earlier work this paper cites.
arXiv:1512.08571
Original
S. Anwar, K. Hwang, W. Sung, Structured pruning of deep convolutional neural networks (2015) · 2015
Earlier work this paper cites.
arXiv:1607.03250
Original
H. Hu, R. Peng, Y.-w. Tai, S. G. Limited, C.-k. Tang, Network Trimming: A Data-Driven Neuron Pruning Approach towards Efficient Deep Architectures (2016) · 2016
Earlier work this paper cites.
A. See, M. T. Luong, C. D. Manning, Compression of neural machine translation models via pruning, CoNLL 2016 - 20th SIGNLL Conference on Computational Natural Language Learning, Proceedings (2016) 291–301 doi:10.18653/v1/k16-1029
2016
Earlier work this paper cites.
H. Li, A. Kadav, I. Durdanovic, H. Samet, H. P. Graf, Pruning Filters for Efficient ConvNets, in: International Conference on Learning Representations, Palais des Congrès Neptune, Toulon, France, 2017, pp. 1–13
2017
Earlier work this paper cites.
arXiv:1511.05077
Original
Z. Mariet, S. Sra, Diversity networks: Neural network compression using determinantal point processes (2017) · 2017
Earlier work this paper cites.

Similar

arXiv:1611.06440
Original
P. Molchanov, S. Tyree, T. Karras, T. Aila, J. Kautz, Pruning convolutional neural networks for resource efficient inference, 5th International Conference on Learning Representations, ICLR 2017 - Conference Track Proceedings (2015) (2017) 1–17 · 2017
Cited alongside, same era.
E. Ameisen, Always start with a stupid model, no exceptions., https://blog.insightdatascience.com/always-start-with-a-stupid-model-no-exceptions-3a22314b9aaa (3 2018)
2018
Cited alongside, same era.
doi:10.1007/978-3-030-00470-5-13
K. Liu, B. Dolan-Gavitt, S. Garg, Fine-pruning: Defending against backdooring attacks on deep neural networks, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 11050 LNCS (2018) 273–294 · 2018
Cited alongside, same era.
arXiv:1802.00124
Original
J. Ye, X. Lu, Z. Lin, J. Z. Wang, Rethinking the smaller-norm-less-informative assumption in channel pruning of convolution layers, arXiv (2017) (2018) 1–11 · 2018
Cited alongside, same era.
doi:10.14722/ndss.2018.23291
Y. Liu, S. Ma, Y. Aafer, W.-C. Lee, J. Zhai, W. Wang, X. Zhang, Trojaning Attack on Neural Networks, in: NDSS, Internet Society, Network and Distributed Systems Security (NDSS) Symposium 2018, San Diego, CA, 2018, pp. 1–15 · 2018
Cited alongside, same era.
H. Siegelmann, Guaranteeing AI Robustness against Deception (GARD), https://www.darpa.mil/program/guaranteeing-ai-robustness-against-deception (2019)
2019
Cited alongside, same era.
X. Xu, Q. Wang, H. Li, N. Borisov, C. A. Gunter, B. Li, Detecting AI Trojans Using Meta Neural Analysis (2019)
2019
Cited alongside, same era.
S. Jha, S. Raj, S. L. Fernandes, S. K. Jha, S. Jha, B. Jalaian, G. Verma, A. Swami, Attribution-based confidence metric for deep neural networks, Advances in Neural Information Processing Systems 32 (NeurIPS) (2019)
2019
Cited alongside, same era.
P. Bajcsy, N. Schaub, M. Majurski, Scientific Calculator for Designing Trojan Detectors in Neural Networks, Association for the Advancement of Artificial Intelligence (AAAI), Fall Symposium Series (FSS), AI in Government and Public Sector Applications (2020) 8
2020
Cited alongside, same era.

Then

IARPA, Intelligence Advanced Research Projects Agency: Trojans in Artificial Intelligence (TrojAI), https://pages.nist.gov/trojai/ (1 2020)
2020
Later among the works it cites.
T. Kulp-McDowall, A. Dima, M. Majurski, TrojAI Literature Review., https://github.com/usnistgov/trojai-literature (12 2020)
2020
Later among the works it cites.
jacobgil, wanglouis49, zepx, eeric, insomnia250, Model Pruning Implementations in GitHub by the listed GitHub users, https://github.com (12 2020)
2020
Later among the works it cites.
NIST, Datasets for Trojans in Artificial Intelligence (TrojAI), https://pages.nist.gov/trojai/ (12 2020)
2020
Later among the works it cites.
Python, Software, Foundation, Pickle - Python object serialization, https://docs.python.org/3/library/index.html (12 2020)
2020
Later among the works it cites.
J. Ellson, E. Gansner, Y. Hu, S. North, Graphviz - Graph Visualization Software, https://graphviz.org/ (12 2020)
2020
Later among the works it cites.
Python, Software, Foundation, PyTorch Conv2d Class, https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html (12 2020)
2020
Later among the works it cites.
VainF-GitHub, Grouped Convolution Issue, https://github.com/VainF/Torch-Pruning/issues/9 (12 2020)
2020
Later among the works it cites.
P. Bajcsy, Implementations of Pruning-Based Trojan Detection in GitHub, https://github.com/usnistgov/trojai-baseline-pruning (1 2021)
2021
Closest in time.

Beyond the bibliography

alphaXiv searches the wider corpus for related work and actual follow-ups.

Open on alphaXiv

alphaXiv is searching for related work…

Reading the bibliography…