Formal Logic-guided Robust Federated Learning against Poisoning Attacks

  • 2024-11-06 02:56:57
  • Dung Thuy Nguyen, Ziyan An, Taylor T. Johnson, Meiyi Ma, Kevin Leach
  • 0

Abstract

Federated Learning (FL) offers a promising solution to the privacy concernsassociated with centralized Machine Learning (ML) by enabling decentralized,collaborative learning. However, FL is vulnerable to various security threats,including poisoning attacks, where adversarial clients manipulate the trainingdata or model updates to degrade overall model performance. Recognizing thisthreat, researchers have focused on developing defense mechanisms to counteractpoisoning attacks in FL systems. However, existing robust FL methodspredominantly focus on computer vision tasks, leaving a gap in addressing theunique challenges of FL with time series data. In this paper, we presentFLORAL, a defense mechanism designed to mitigate poisoning attacks in federatedlearning for time-series tasks, even in scenarios with heterogeneous clientdata and a large number of adversarial participants. Unlike traditionalmodel-centric defenses, FLORAL leverages logical reasoning to evaluate clienttrustworthiness by aligning their predictions with global time-series patterns,rather than relying solely on the similarity of client updates. Our approachextracts logical reasoning properties from clients, then hierarchically infersglobal properties, and uses these to verify client updates. Through formallogic verification, we assess the robustness of each client contribution,identifying deviations indicative of adversarial behavior. Experimental resultson two datasets demonstrate the superior performance of our approach comparedto existing baseline methods, highlighting its potential to enhance therobustness of FL to time series applications. Notably, FLORAL reduced theprediction error by 93.27% in the best-case scenario compared to thesecond-best baseline. Our code is available athttps://anonymous.4open.science/r/FLORAL-Robust-FTS.

 

Quick Read (beta)

loading the full paper ...