Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
156 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
45 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Efficient DNN-Powered Software with Fair Sparse Models (2407.02805v1)

Published 3 Jul 2024 in cs.SE and cs.AI

Abstract: With the emergence of the Software 3.0 era, there is a growing trend of compressing and integrating large models into software systems, with significant societal implications. Regrettably, in numerous instances, model compression techniques impact the fairness performance of these models and thus the ethical behavior of DNN-powered software. One of the most notable example is the Lottery Ticket Hypothesis (LTH), a prevailing model pruning approach. This paper demonstrates that fairness issue of LTHbased pruning arises from both its subnetwork selection and training procedures, highlighting the inadequacy of existing remedies. To address this, we propose a novel pruning framework, Ballot, which employs a novel conflict-detection-based subnetwork selection to find accurate and fair subnetworks, coupled with a refined training process to attain a high-performance model, thereby improving the fairness of DNN-powered software. By means of this procedure, Ballot improves the fairness of pruning by 38.00%, 33.91%, 17.96%, and 35.82% compared to state-of-the-art baselines, namely Magnitude Pruning, Standard LTH, SafeCompress, and FairScratch respectively, based on our evaluation of five popular datasets and three widely used models. Our code is available at https://anonymous.4open.science/r/Ballot-506E.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (97)
  1. [n. d.]. Anonymized Repository - Anonymous GitHub. https://anonymous.4open.science/r/Ballot-506E.
  2. [n. d.]. Rasbt/Deeplearning-Models: A Collection of Various Deep Learning Architectures, Models, and Tips. https://github.com/rasbt/deeplearning-models/tree/master.
  3. [n. d.]. Torch.Nn — PyTorch 2.1 Documentation. https://pytorch.org/docs/stable/nn.html#module-torch.nn.utils.
  4. 2020. CIFAR-100 Datasets. https://www.cs.toronto.edu/~kriz/cifar.html. https://www.cs.toronto.edu/~kriz/cifar.html.
  5. TensorFlow: A System for Large-Scale Machine Learning. In 12th USENIX Symposium on Operating Systems Design and Implementation (OSDI’16). 265–283.
  6. One-Network Adversarial Fairness. Proc. of AAAI (2019), 2412–2420. https://doi.org/10.1609/aaai.v33i01.33012412
  7. Automated Test Generation to Detect Individual Discrimination in AI Models. arXiv preprint arXiv:1809.03260 (2018). arXiv:1809.03260
  8. Themis: Automatically Testing Software for Discrimination. In Proceedings of the 2018 26th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering - ESEC/FSE 2018. 871–875. https://doi.org/10.1145/3236024.3264590
  9. Biasfinder: Metamorphic test generation to uncover bias for sentiment analysis systems. IEEE Transactions on Software Engineering (2021), 5087–5101.
  10. Fatma Başak Aydemir and Fabiano Dalpiaz. 2018. A Roadmap for Ethics-Aware Software Engineering. In Proceedings of the International Workshop on Software Fairness. 15–21. https://doi.org/10.1145/3194770.3194778
  11. Jimmy Ba and Rich Caruana. 2014. Do deep nets really need to be deep? Proc. of NeurIPS (2014).
  12. Representation learning: A review and new perspectives. IEEE transactions on pattern analysis and machine intelligence (2013), 1798–1828.
  13. Robustness may be at odds with fairness: An empirical study on class-wise accuracy. In NeurIPS 2020 Workshop on pre-registration in machine learning. 325–342.
  14. Data Decisions and Theoretical Implications When Adversarially Learning Fair Representations. arXiv:1707.00075 [cs] (July 2017). arXiv:1707.00075 [cs]
  15. Simon says: Evaluating and mitigating bias in pruned neural networks with knowledge distillation. arXiv preprint arXiv:2106.07849 (2021).
  16. Demographic dialectal variation in social media: A case study of African-American English. arXiv preprint arXiv:1608.08868 (2016).
  17. Tim Brennan and William L. Oliver. 2013. Emergence of Machine Learning Techniques in Criminology: Implications of Complexity in Our Data and in Research Questions. Criminology & Pub. Pol’y (2013), 551.
  18. Language Models Are Few-Shot Learners. https://doi.org/10.48550/arXiv.2005.14165 arXiv:2005.14165 [cs]
  19. Yuriy Brun and Alexandra Meliou. 2018. Software Fairness. In Proceedings of the 2018 26th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering. 754–759. https://doi.org/10.1145/3236024.3264838
  20. T. Anne Cleary. 1966. Test Bias: Validity of the Scholastic Aptitude Test for Negro and White Students in Integrated Colleges. ETS Research Bulletin Series (1966), i–23.
  21. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805 (2018).
  22. Fairness through Awareness. In Proceedings of the 3rd Innovations in Theoretical Computer Science Conference on - ITCS ’12. 214–226. https://doi.org/10.1145/2090236.2090255
  23. Explanation-guided fairness testing through genetic algorithm. In Proc. of ICSE. 871–882.
  24. Certifying and Removing Disparate Impact. In Proc. of KDD. 259–268. https://doi.org/10.1145/2783258.2783311
  25. Jonathan Frankle and Michael Carbin. 2018. The lottery ticket hypothesis: Finding sparse, trainable neural networks. arXiv preprint arXiv:1803.03635 (2018).
  26. Jonathan Frankle and Michael Carbin. 2019. The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks. In Proc. of ICLR.
  27. Linear mode connectivity and the lottery ticket hypothesis. In Proc. of ICML. 3259–3269.
  28. Fairness Testing: Testing Software for Discrimination. In Proceedings of the 2017 11th Joint Meeting on Foundations of Software Engineering. 498–510. https://doi.org/10.1145/3106237.3106277
  29. Xuanqi Gao. 2022. FairNeuron.
  30. Fairneuron: Improving Deep Neural Network Fairness with Adversary Games on Selective Neurons. In Proc. of ICSE. 921–933. https://doi.org/10.1145/3510003.3510087
  31. CILIATE: Towards Fairer Class-based Incremental Learning by Dataset and Training Refinement. arXiv preprint arXiv:2304.04222 (2023).
  32. Shortcut Learning in Deep Neural Networks. Nature Machine Intelligence 2, 11 (November 2020), 665–673. https://doi.org/10.1038/s42256-020-00257-z arXiv:2004.07780 [cs, q-bio]
  33. Robert M. Guion. 1966. Employment Tests and Discriminatory Hiring. Industrial Relations: A Journal of Economy and Society (1966), 20–37.
  34. Antonio Gulli and Sujit Pal. 2017. Deep learning with Keras. Packt Publishing Ltd.
  35. FairRec: Fairness Testing for Deep Recommender Systems. arXiv preprint arXiv:2304.07030 (2023).
  36. Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding. arXiv preprint arXiv:1510.00149 (2015).
  37. Learning both Weights and Connections for Efficient Neural Network. In Proc. of NeurIPS. 1135–1143.
  38. Equality of Opportunity in Supervised Learning. Proc. of NeurIPS (2016), 3315–3323.
  39. Deep Residual Learning for Image Recognition. In Proc. of CVPR. 770–778.
  40. Filter pruning via geometric median for deep convolutional neural networks acceleration. In Proc. of CVPR. 4340–4349.
  41. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531 (2015).
  42. Labeled Faces in the Wild: A Database for Studying Face Recognition in Unconstrained Environments. Technical Report.
  43. Quantization and training of neural networks for efficient integer-arithmetic-only inference. In Proc. of CVPR. 2704–2713.
  44. Tian Jin. 2022. On neural network pruning’s effect on generalization. Ph. D. Dissertation.
  45. Faisal Kamiran and Toon Calders. 2009. Classifying without Discriminating. In 2009 2nd International Conference on Computer, Control and Communication. 1–6.
  46. Faisal Kamiran and Toon Calders. 2012. Data Preprocessing Techniques for Classification without Discrimination. Knowledge and Information Systems (2012), 1–33. https://doi.org/10.1007/s10115-011-0463-8
  47. Attribute and simile classifiers for face verification. In Proc. of ICCV. 365–372.
  48. Ya Le and Xuan Yang. 2015. Tiny imagenet visual recognition challenge. CS 231N (2015), 3.
  49. Optimal brain damage. Proc. of NeurIPS (1989).
  50. Pruning filters for efficient convnets. arXiv preprint arXiv:1608.08710 (2016).
  51. Training Data Debugging for the Fairness of Machine Learning Software. In Proc. of ICSE. 2215–2227. https://doi.org/10.1145/3510003.3510091
  52. Can Pruning Improve Certified Robustness of Neural Networks? Transactions on Machine Learning Research (2023).
  53. Pruning and quantization for deep neural network acceleration: A survey. Neurocomputing (2021), 370–403.
  54. On-Demand Deep Model Compression for Mobile Devices: A Usage-Driven Model Selection Framework. In Proceedings of the 16th Annual International Conference on Mobile Systems, Applications, and Services. 389–400. https://doi.org/10.1145/3210240.3210337
  55. Deep Learning Face Attributes in the Wild. In Proc. of ICCV.
  56. Estimating the carbon footprint of bloom, a 176b parameter language model. arXiv preprint arXiv:2211.02001 (2022).
  57. Jian-Hao Luo and Jianxin Wu. 2017. An entropy-based pruning method for cnn compression. arXiv preprint arXiv:1706.05791 (2017).
  58. Sanity checks for lottery tickets: Does your winning ticket really win the jackpot? Proc. of NeurIPS (2021), 12749–12760.
  59. Sanity Checks for Lottery Tickets: Does Your Winning Ticket Really Win the Jackpot?. In Proc. of NeurIPS. 12749–12760.
  60. Software Engineering for AI-Based Systems: A Survey. ACM Transactions on Software Engineering and Methodology (2022), 37e:1–37e:59. https://doi.org/10.1145/3487043
  61. A Survey on Deep Neural Network Compression: Challenges, Overview, and Solutions. arXiv:2010.03954 [cs, eess]
  62. Importance estimation for neural network pruning. In Proc. of CVPR. 11264–11272.
  63. Deep Model Compression for Mobile Platforms: A Survey. Tsinghua Science and Technology (2019), 677–693. https://doi.org/10.26599/TST.2018.9010103
  64. Wrapnet: Neural net inference with ultra-low-resolution arithmetic. arXiv preprint arXiv:2007.13242 (2020).
  65. Cathy O’neil. 2016. Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy. Crown.
  66. PaddlePaddle. [n. d.]. prune model-API Document.
  67. Michela Paganini. 2020. Prune responsibly. arXiv preprint arXiv:2009.09936 (2020).
  68. On Fairness and Calibration. (n. d.), 10.
  69. Adam Polyak and Lior Wolf. 2015. Channel-level acceleration of deep face representations. IEEE Access (2015), 2163–2175.
  70. Mitigating Bias in Algorithmic Hiring: Evaluating Claims and Practices. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency. 469–481.
  71. Fairness GAN: Generating Datasets with Fairness Properties Using a Generative Adversarial Network. IBM Journal of Research and Development (2019), 3–1.
  72. Green ai. Commun. ACM (2020), 54–63.
  73. Grad-cam: Visual explanations from deep networks via gradient-based localization. In Proc. of ICCV. 618–626.
  74. Agam Shah. 2023. Nvidia CEO Huang: Get Ready for Software 3.0.
  75. Compressing pre-trained models of code into 3 mb. In Proceedings of the 37th IEEE/ACM International Conference on Automated Software Engineering. 1–12.
  76. Karen Simonyan and Andrew Zisserman. 2014. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556 (2014).
  77. Causality-based neural network repair. In Proc. of ICSE. 338–349.
  78. Degree-quant: Quantization-aware training for graph neural networks. arXiv preprint arXiv:2008.05000 (2020).
  79. Fair Scratch Tickets: Finding Fair Sparse Networks Without Weight Training. In Proc. of CVPR. 24406–24416.
  80. TensorflowBlog. [n. d.]. TensorFlow Model Optimization Toolkit — Pruning API.
  81. Image fairness in deep learning: problems, models, and challenges. Neural Computing and Applications (2022), 12875–12893.
  82. Analysis and Applications of Class-Wise Robustness in Adversarial Training. In Proc. of KDD. 1561–1570.
  83. PyTorch Tutorials. [n. d.]. Pruning Tutorial.
  84. Automated Directed Fairness Testing. In Proceedings of the 33rd ACM/IEEE International Conference on Automated Software Engineering - ASE 2018. 98–108. https://doi.org/10.1145/3238147.3238165
  85. Hiring Algorithms: An Ethnography of Fairness in Practice. (2019).
  86. Ana Ware. 2022. How Giant AI Workloads and the Looming “Bandwidth Wall” Are Impacting System Architectures.
  87. Fairprune: Achieving fairness through pruning for dermatological disease diagnosis. In International Conference on Medical Image Computing and Computer-Assisted Intervention. 743–753.
  88. Diffchaser: Detecting disagreements for deep neural networks.
  89. Fairgan: Fairness-aware Generative Adversarial Networks. In 2018 IEEE International Conference on Big Data (Big Data). 570–575.
  90. Biasrv: Uncovering biased sentiment predictions at runtime. In Proceedings of the 29th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering. 1540–1544.
  91. Mitigating Unwanted Biases with Adversarial Learning. In Proc. of AAAI. 335–340. https://doi.org/10.1145/3278721.3278779
  92. Achieving Non-Discrimination in Data Release. In Proc. of KDD. 1335–1344. https://doi.org/10.1145/3097983.3098167
  93. Efficient white-box fairness testing through gradient search. In Proc. of ISSTA. 103–114.
  94. White-Box Fairness Testing through Adversarial Sampling. In Proc. of ICSE. 949–960. https://doi.org/10.1145/3377811.3380331
  95. QVIP: an ILP-based formal verification approach for quantized neural networks. In Proceedings of the 37th IEEE/ACM International Conference on Automated Software Engineering. 1–13.
  96. A Survey of Large Language Models. https://doi.org/10.48550/arXiv.2303.18223 arXiv:2303.18223 [cs]
  97. Safety and Performance, Why not Both? Bi-Objective Optimized Model Compression toward AI Software Deployment. In Proceedings of the 37th IEEE/ACM International Conference on Automated Software Engineering. 1–13.

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now
X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets