RankGuard is a decentralized OLTR system that filters model updates using local click data for poisoning resistance and supplies the first formal convergence guarantee for decentralized OLTR.
Byzantine-robust decentralized learning via self-centered clipping
7 Pith papers cite this work. Polarity classification is still indexing.
7
Pith papers citing it
citation-role summary
background 2
citation-polarity summary
roles
background 2polarities
background 2representative citing papers
An accelerated one-step-delayed stochastic method achieves the centralized O(1/sqrt(N)) rate with M up to sqrt(rho) N^{3/4} workers and a matching lower bound for linear-span first-order methods.
DRSGD-ByMI identifies Byzantine machines via sample-splitting score statistics with FDR control, then prunes them to recover sufficient connectivity and achieve order-optimal convergence rates identical to standard decentralized SGD.
GT-PD achieves linear convergence to a variance-determined neighborhood in Byzantine settings by clipping messages and using dual-metric probabilistic dropout to preserve doubly stochastic mixing; GT-PD-L adds leaky integration for partial isolation.
SketchGuard decouples Byzantine filtering from aggregation in decentralized federated learning by exchanging k-dimensional Count Sketches for screening and full models only from accepted neighbors, achieving up to 50-70% communication savings while proving convergence and matching SOTA robustness.
A single global merge at the final step of decentralized SGD matches the convergence rate of parallel SGD while improving test accuracy under high data heterogeneity.
RESIST achieves algorithmic and statistical convergence guarantees for strongly convex, PL, and nonconvex ERM under MITM attacks via multistep consensus gradient descent plus robust screening.
citing papers explorer
-
Efficient and Robust Online Learning to Rank in Decentralized Systems
RankGuard is a decentralized OLTR system that filters model updates using local click data for poisoning resistance and supplies the first formal convergence guarantee for decentralized OLTR.
-
Near-Optimal Decentralized Stochastic Convex Optimization over Networks
An accelerated one-step-delayed stochastic method achieves the centralized O(1/sqrt(N)) rate with M up to sqrt(rho) N^{3/4} workers and a matching lower bound for linear-span first-order methods.
-
Toward Exact Convergence in Byzantine-Robust Decentralized Learning: A Statistical Identification Approach
DRSGD-ByMI identifies Byzantine machines via sample-splitting score statistics with FDR control, then prunes them to recover sufficient connectivity and achieve order-optimal convergence rates identical to standard decentralized SGD.
-
Convergence of Byzantine-Resilient Gradient Tracking via Probabilistic Edge Dropout
GT-PD achieves linear convergence to a variance-determined neighborhood in Byzantine settings by clipping messages and using dual-metric probabilistic dropout to preserve doubly stochastic mixing; GT-PD-L adds leaky integration for partial isolation.
-
SketchGuard: Scaling Byzantine-Robust Decentralized Federated Learning via Sketch-Based Screening
SketchGuard decouples Byzantine filtering from aggregation in decentralized federated learning by exchanging k-dimensional Count Sketches for screening and full models only from accepted neighbors, achieving up to 50-70% communication savings while proving convergence and matching SOTA robustness.
-
On the Surprising Effectiveness of a Single Global Merging in Decentralized Learning
A single global merge at the final step of decentralized SGD matches the convergence rate of parallel SGD while improving test accuracy under high data heterogeneity.
-
RESIST: Resilient Decentralized Learning Using Consensus Gradient Descent
RESIST achieves algorithmic and statistical convergence guarantees for strongly convex, PL, and nonconvex ERM under MITM attacks via multistep consensus gradient descent plus robust screening.