Targeted Attacks on Timeseries Forecasting

Yuvaraj Govindarajulu, Avinash Amballa, Pavan Kulkarni, Manojkumar Parmar

Keywords: timeseries, forecasting, targeted attacks, adversarial ml, ai security, apgd

Abstract: Real-world deep learning models built for Time Series Forecasting are used in several critical applications from medical devices to the security domain. Many previous works have shown how deep learning models are prone to adversarial attacks and studied their vulnerabilities. However, the vulnerabilities of time series models for forecasting due to adversarial inputs are not extensively studied. While attack on a forecasting model might be intended to deteriorate the performance of the model, it is more effective, if the attack is focused on a specific impact on the model's output. In this paper, we propose a novel formulation of Directional, Amplitudinal, and Temporal targeted adversarial attacks on time series forecasting models. These targeted attacks create a specific impact or hide a potential high-impact area on the forecasting output. We use the existing adversarial attack techniques from the computer vision domain and adapt them for time series. Additionally, we propose a modified version of the Auto Projected Gradient Descent attack for targeted attacks. We explore the impact of the proposed targeted attacks against untargeted attacks. We use KS-Tests to statistically prove the impact of the attack. Our experimental results demonstrate how targeted attacks on time series models are practical and are more powerful in terms of statistical similarity. It is, hence difficult to detect through statistical methods. We believe that this work opens a new paradigm in the time series forecasting domain and is an important consideration for developing better defenses.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Social Aspects of Machine Learning (eg, AI safety, fairness, privacy, interpretability, human-AI interaction, ethics)