MultiKrum: Revolutionizing Distributed Learning with Enhanced Robustness

This research introduces MultiKrum, a significant advancement in distributed learning, providing the first theoretical guarantees for its robustness. It also unveils a novel concept, the optimal robustness coefficient, to quantify the accuracy of mean estimation in the face of adversaries. This work demonstrates impressive progress in creating more reliable and efficient aggregation rules.

• •MultiKrum, a practical aggregation rule, now has its robustness proven.
• •The paper introduces a new 'optimal robustness coefficient' for more precise accuracy measurement.
• •The research improves robustness bounds compared to the Krum rule in realistic scenarios.