mlsanityData to 14 Oct 2026

HARMONY: Bridging the Personalization-Generalization Gap by Mitigating Representation Skew in Heterogeneous Split Federated Learning

arXiv 2605.07211 ↗First posted 8 May 2026cs.LG cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

Mobile devices face diverse resource constraints and non-IID data class distributions, requiring fast on-device inference for local in-distribution (ID) classes and on-demand remote support for client-specific out-of-distribution (OOD) classes. Hybrid split federated learning (Hybrid SFL) couples personalized client-side front ends (supporting early exit) with a generalized server-side backend for fallback inference, balancing accuracy and cost. However, under client architectural heterogeneity, the existing hybrid SFL suffers from representation skew, where features from customized extractors fail to align in the shared space, leading to a sharp degradation in the server model responsible for OOD prediction. We propose HARMONY, the first hybrid SFL framework to support heterogeneous client architectures. HARMONY modifies meta-learning to simulate diverse extractors across parameters and architectures, and to learn to personalize. To mitigate representation skew, HARMONY conducts server-side contrastive learning to align extracted features, neither sacrificing clients' personalization nor sharing raw labels. Compared to the state of the art across multiple datasets and model families, HARMONY improves test accuracy by up to 43.0%/28.3% without/with OOD, respectively, while maintaining acceptable latency.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 28 of its 40 references that are in the map. First posted on arXiv 8 May 2026.

Builds on

28 of 40 references.

References · 28

  1. 2016Communication-Efficient Learning of Deep Networks from Decentralized Datacited by 7,669
  2. 2017Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networkscited by 4,897
  3. 2017Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithmscited by 4,349
  4. 2018Federated Optimization in Heterogeneous Networkscited by 2,987
  5. 2019Federated Learning: Challenges, Methods, and Future Directionscited by 1,538
  6. 2020Adaptive Federated Optimizationcited by 1,046
  7. 2018On First-Order Meta-Learning Algorithmscited by 1,018
  8. 2020Tackling the Objective Inconsistency Problem in Heterogeneous Federated Optimizationcited by 906
  9. 2020Ensemble Distillation for Robust Model Fusion in Federated Learningcited by 682
  10. 2019A Theoretical Analysis of Contrastive Unsupervised Representation Learningcited by 530
  11. 2019FedMD: Heterogenous Federated Learning via Model Distillationcited by 459
  12. 2018Split learning for health: Distributed deep learning without sharing raw patient datacited by 395
16 more
  1. 2019FedPAQ: A Communication-Efficient Federated Learning Method with Periodic Averaging and Quantizationcited by 356
  2. 2021FedProto: Federated Prototype Learning across Heterogeneous Clientscited by 352
  3. 2020HeteroFL: Computation and Communication Efficient Federated Learning for Heterogeneous Clientscited by 337
  4. 2017Unified Deep Supervised Domain Adaptation and Generalizationcited by 323
  5. 2020SplitFed: When Federated Learning Meets Split Learningcited by 310
  6. 2018Distributed learning of deep neural network over multiple agentscited by 279
  7. 2023Heterogeneous Federated Learning: State-of-the-art and Research Challengescited by 161
  8. 2024AdaptSFL: Adaptive Split Federated Learning in Resource-constrained Edge Networkscited by 64
  9. 2022Test-Time Robust Personalization for Federated Learningcited by 39
  10. 2021Simpler, Faster, Stronger: Breaking The log-K Curse On Contrastive Learners With FlatNCEcited by 22
  11. 2024FIARSE: Model-Heterogeneous Federated Learning via Importance-Aware Submodel Extractioncited by 21
  12. 2024Personalized Federated Learning via Feature Distribution Adaptationcited by 13
  13. 2025Federated Learning with Domain Shift Erasercited by 7
  14. 2025FSL-SAGE: Accelerating Federated Split Learning via Smashed Activation Gradient Estimationcited by 7
  15. 2025Feature Distillation is the Better Choice for Model-Heterogeneous Federated Learningcited by 5
  16. 2025FedFetch: Faster Federated Learning with Adaptive Downstream Prefetchingcited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025FedeCouple: Fine-Grained Balancing of Global-Generalization and Local-Adaptability in Federated Learning
  2. 2025Towards a Larger Model via One-Shot Federated Learning on Heterogeneous Client Models
  3. 2023Fed-QSSL: A Framework for Personalized Federated Learning under Bitwidth and Data Heterogeneity
  4. 2026FedLTLib: A Comprehensive Benchmark for Federated Long-Tail Learning
  5. 2026COSMOS: Model-Agnostic Personalized Federated Learning with Clustered Server Models and Pseudo-Label-Only Communication
  6. 2021Heterogeneous Federated Learning using Dynamic Model Pruning and Adaptive Gradient
  7. 2025Co-LoRA: Collaborative Model Personalization on Heterogeneous Multi-Modal Clients
  8. 2024Towards Diverse Device Heterogeneous Federated Learning via Task Arithmetic Knowledge Integration