Back to feed
arXiv cs.LG
arXiv cs.LG
7/13/2026
HERO: A Heterogeneity-Aware Benchmark Library for Federated Continual Learning

HERO: A Heterogeneity-Aware Benchmark Library for Federated Continual Learning

Short summary

HERO is a heterogeneity-aware benchmark library for federated continual learning that decouples task splits, client data splits, and client task sequences to enable fair comparisons. It introduces controllable parameters for data skew and task-order mismatch, revealing that method behavior changes across easy and heterogeneous settings and that average accuracy can mask weak bottom-client performance. The library includes benchmark streams, method implementations, and reporting scripts for reproducible evaluation.

  • •Decouples task split, client data split, and task sequence for fair FCL comparison
  • •Reveals average accuracy hides weak bottom-client performance in heterogeneous settings
  • •Includes graph-based Domain-IL case study on OGB-MolPCBA beyond image tasks

Generated with AI, which can make mistakes.

#open-source#research-breakthrough
Read full article at arXiv cs.LG

Is this a good recommendation for you?

Comments

Failed to load comments. Please try again.

Explore more

SUM: Unified Geometric Surgery on Spatio-Temporal Adaptation Vectors for Federated Class Incremental Learning
arXiv cs.LG
FedCC: A Low-Resource Federated Adaptation of Foundation Models for Robust Corpus Callosum localization in Fetal Ultrasound Images
arXiv cs.LG
Skyfall AI launches MORPHEUS: a persistent enterprise simulation benchmark for continual reinforcement learning
MarkTechPost
ReCoLoRA: Spectrum-Aware Recursive Consolidation for Continual LLM Fine-Tuning
arXiv cs.LG