电磁频谱预测的智能对抗方法、系统、设备及介质
By employing both traditional and triggered poisoning attack methods in spectrum prediction, the vulnerability of deep learning models to poisoning attacks is addressed, enabling flexible and covert attacks on spectrum prediction models and enhancing the security and defense capabilities of spectrum prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIDIAN UNIV
- Filing Date
- 2023-11-28
- Publication Date
- 2026-07-17
AI Technical Summary
Existing deep learning-based spectrum prediction techniques are vulnerable to poisoning attacks, and research mainly focuses on the overall cognitive spectrum decision-making process, lacking in-depth analysis of time series prediction and exploration of flexible attack methods.
This paper proposes an optimization problem for poisoning attacks based on time series prediction neural networks. It adopts both traditional and trigger-based poisoning attack methods, and achieves the attack on the spectrum prediction model by adding specific increments or implanting backdoors to the initial time slots.
It effectively reduces the cost of modifying the training set, increases the flexibility and stealth of attacks, and improves the security and defense capabilities of the spectrum prediction model.
Smart Images
Figure CN117641364B_ABST