Watson_CS592_ProjectProposal

START

Securing Time-Series Models with unlearning

Project Definition

Among many concerns, the consequences of reverse-engineering algorithms to uncover sensitive data have been a primary driver behind the model unlearning movement [1]. Although current model unlearning strategies have been primarily concentrated on image and text data [2], time-series reverse-engineering poses a great risk to a variety of domains including healthcare, networking, and finance.

The proposed project will explore the concept of unlearning sensitive location data used for time series forecasting to address the feasibility of unlearning information used in time-series algorithms.

Existing Research

• Ye and Lu (2024) propose an unlearning strategy for sequential unlearning systems to privatize these systems and remove ”specific client information” [7].
• Du, Chen, et al. (2019) present an unlearning method for anomaly detection algorithms to improve performance, particularly in how systems approach false positives and false negatives [8].
• Fan(2023) proposes a model unlearning technique specifically oriented towards IoT (Internet of Things) anomaly detection models to address growing industrial security concerns [9].

While there have been many developments in the field of unlearning, many limitations still exist...

Limitations

• While a few studies focused on time-series models exist, they represent a minority of the existing research despite the prevalence of the models.

• Comparable studies generally address anomaly detection, which is just one of many time-series-centered models.

• Some papers identify problems with exploding loss and catastrophic forgetting in the unlearning process [8].
• Scalability has been a universal concern for model unlearning, as the process historically does not scale well for models trained on large datasets [10].

Proposed Methods and Outputs

Outputs

• Forecast Model
• Unlearned Model
• Performance Metrics
• MAE and RSME for base model and unlearned model
• Proof of Unlearning
• Accuracy of the regression model
• XAI Output
• Documentation & Code

Methods

• Model Training: Forecast model built on time-series location data (LSTM, Time-Series Transformer, XGBoost)
• Model Unlearning: Unlearn location using pruning, fine-tuning, and masking
• Performance Evaluations: MAE and RSME
• Unlearning Evaluations: Use a regression model to reconstruct location data, XAI analysis

10/20-11/2 - Model Unlearning Implementation

10/10-10-19 - Model Training and Initial Evaluation

10/1-10/9 - Data Collection and Preprocessing

9/27-9/30- Project Proposal

Project Timeline

11/10-12/2: - Final Presentation Preparation - Write Project Report - Final Presentation

12/9- Project Report Due

11/3-11/9- Validation and Tetsing

Thank You! Questions?

References

[1] H. Liu, P. Xiong, T. Zhu, and P. S. Yu, “A Survey on Machine Unlearning: Techniques and New Emerged Privacy Risks,” Jun. 2024, arXiv:2406.06186 [cs]. [Online]. Available: http://arxiv.org/abs/2406.06186 [2] W. Wang, Z. Tian, C. Zhang, and S. Yu, “Machine Unlearning: A Comprehensive Survey,” Jul. 2024, arXiv:2405.07406 [cs]. [Online]. Available: http://arxiv.org/abs/2405.07406 [3] “Data from fitness app Strava highlights locations of soldiers, U.S. bases - CBS News.” [Online]. Available: https://www.cbsnews.com/news/fitness-devices-soldiers-sensitive-military-bases-location-report/ [4] K. Childs, D. Nolting, and A. Das, “Heat Marks the Spot: De-Anonymizing Users’ Geographical Data on the Strava Heatmap.” [5] “Moving Time, Speed, and Pace Calculations,” Oct. 2023. [Online]. Available: https://support.strava.com/hc/en-us/articles/115001188684-Moving-Time-Speed-and-Pace-Calculations [6] Y. Qu, X. Yuan, M. Ding, W. Ni, T. Rakotoarivelo, and D. Smith, “Learn to Unlearn: A Survey on Machine Unlearning,” Oct. 2023, arXiv:2305.07512 [cs]. [Online]. Available: http://arxiv.org/abs/2305.07512

[7] S. Ye and J. Lu, “Sequence Unlearning for Sequential Recommender Systems,” in AI 2023: Advances in Artificial Intelligence, T. Liu, G. Webb, L. Yue, and D. Wang, Eds. Singapore: Springer Nature, 2024, pp. 403–415. [8] M. Du, Z. Chen, C. Liu, R. Oak, and D. Song, “Lifelong Anomaly Detection Through Unlearning,” in Proceedings of the 2019 ACM SIGSAC Conference on Computer and Communications Security, ser. CCS ’19. New York, NY, USA: Association for Computing Machinery, Nov. 2019, pp. 1283–1297. [Online]. Available: https://dl.acm.org/doi/10.1145/3319535.3363226 [9] Jiamin Fan, “Machine Learning and Unlearning for IoT Anomaly Detection,” Dissertation, University of Vic- toria, 2023. [10] J. Xu, Z. Wu, C. Wang, and X. Jia, “Machine Unlearning: Solutions and Challenges,” IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 8, no. 3, pp. 2150–2168, Jun. 2024, conference Name: IEEE Transactions on Emerging Topics in Computational Intelligence. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/10488864