Byzantine Agreement with Predictions

2025-05-03

AI Summary

This paper studies Byzantine consensus with predictions: each process may attach a potentially erroneous classification prediction (e.g., output from safety monitoring) to its proposal, aiming to accelerate consensus when predictions are accurate and gracefully degrade to classical optimal performance when inaccurate. We formally define this model for the first time and prove that the message complexity lower bound remains Ω(n²), while predictions strictly improve time complexity. We propose an optimal, prediction-quality-adaptive algorithm: it achieves sublinear-round consensus under accurate predictions and degrades to O(f) rounds in the worst case—matching the round complexity of optimal prediction-free protocols. We further provide a tight lower bound proving that this time complexity is asymptotically optimal. Our core contribution is establishing the first theoretical framework for prediction-augmented consensus, enabling principled trade-offs between prediction accuracy and robustness.

Technology Category

• Reasoning under Uncertainty: Stochastic Optimization
• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories
• Machine Learning: Online Learning & Bandits

Application Category

• Social Networks and Social Media: Influence propagation, information diffusion, and the prediction on networks
• Security and Privacy: Large-scale security measurements
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

In this paper, we study the problem of emph{Byzantine Agreement with predictions}. Along with a proposal, each process is also given a prediction, i.e., extra information which is not guaranteed to be true. For example, one might imagine that the prediction is produced by a network security monitoring service that looks for patterns of malicious behavior. Our goal is to design an algorithm that is more efficient when the predictions are accurate, degrades in performance as predictions decrease in accuracy, and still in the worst case performs as well as any algorithm without predictions even when the predictions are completely inaccurate. On the negative side, we show that Byzantine Agreement with predictions still requires $Omega(n^2)$ messages, even in executions where the predictions are completely accurate. On the positive side, we show that emph{classification predictions} can help improve the time complexity. For (synchronous) Byzantine Agreement with classification predictions, we present new algorithms that leverage predictions to yield better time complexity, and we show that the time complexity achieved is optimal as a function of the prediction quality.

Problem

• Study Byzantine Agreement with unreliable process predictions
• Design algorithm adapting efficiency to prediction accuracy
• Prove message complexity limits despite accurate predictions

Innovation

• Leveraging predictions for Byzantine Agreement efficiency
• Optimal time complexity with classification predictions
• Maintaining worst-case performance without predictions