Researchers have developed a federated learning approach to train aircraft engine prognostics models while addressing both benign and adversarial data heterogeneity. The study utilized a multi-task one-dimensional convolutional neural network on the C-MAPSS benchmark to evaluate methods for benign heterogeneity, finding that shared-representation personalization significantly improved model accuracy. For adversarial scenarios, a backdoor attack demonstrated a high success rate against standard averaging, highlighting the need for explicit safety evaluations. The Krum aggregation method proved effective in reducing attack success and withstanding coordinated attackers, especially when combined with personalization, achieving robust performance with minimal accuracy loss. AI
IMPACT Enhances the security and accuracy of AI models in critical infrastructure by addressing data heterogeneity and adversarial attacks.
RANK_REASON Academic paper detailing a novel approach to federated learning for a specific application.
- C-MAPSS
- federated learning
- Krum
- One-dimensional convolutional neural network (1D-CNN) image reconstruction for electrical impedance tomography
- Remaining Useful Life Prediction of Rolling Bearings Using PSR, JADE, and Extreme Learning Machine
- Commercial Modular Aero-Propulsion System Simulation
- Hugging Face
- multi-task one-dimensional convolutional neural network
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
