SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2603.24209 · 2026

HEART-PFL: Stable Personalized Federated Learning under Heterogeneity with Hierarchical Directional Alignment and Adversarial Knowledge Transfer

Minjun Kim, Minje Kim

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 13 functions out of this paper's own repositories and ran 10 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
danny0628/HEART-PFL canonical 10 of 13
FunctionStatusWhere it lives
FedAveraging Ran danny0628/HEART-PFL/aggregation/averaging.py
pointer only (licence: NOASSERTION) · get_code("0aaece534c92f75d")
adaad_inner_loss Ran danny0628/HEART-PFL/aggregation/AT_helper.py
pointer only (licence: NOASSERTION) · get_code("f6a6a32c98350e2b")
computeAUROC Ran danny0628/HEART-PFL/aggregation/distillation.py
pointer only (licence: NOASSERTION) · get_code("3975b1aa54ba551d")
conv1x1 Ran danny0628/HEART-PFL/models/resnet_bn_utils.py
pointer only (licence: NONE) · get_code("2a80220dabcb742a")
conv3x3 Ran danny0628/HEART-PFL/models/resnet_bn_utils.py
pointer only (licence: NOASSERTION) · get_code("54ef4a4d895678ee")
create_batch_norm Ran danny0628/HEART-PFL/models/resnet_bn_utils.py
pointer only (licence: NOASSERTION) · get_code("b91186f3582c95f3")
divergence Ran danny0628/HEART-PFL/aggregation/distillation.py
pointer only (licence: NOASSERTION) · get_code("5405f5e2e7253079")
flower102_noniid Ran danny0628/HEART-PFL/datasets/prepare_data.py
pointer only (licence: NOASSERTION) · get_code("ae97cecf56042ffb")
mtad_inner_loss Ran danny0628/HEART-PFL/aggregation/AT_helper.py
pointer only (licence: NOASSERTION) · get_code("67a43807986ec3bc")
test Ran danny0628/HEART-PFL/aggregation/distillation.py
pointer only (licence: NOASSERTION) · get_code("6742d7a324b62056")
Madry_PGD Not yet run danny0628/HEART-PFL/aggregation/AT_helper.py
pointer only (licence: NOASSERTION) · get_code("750f12ea8853901e")
cifar_noniid Not yet run danny0628/HEART-PFL/datasets/prepare_data.py
pointer only (licence: NOASSERTION) · get_code("51c693c3a5f724fe")
get_dataset Not yet run danny0628/HEART-PFL/datasets/prepare_data.py
pointer only (licence: NOASSERTION) · get_code("6fcb5d0981f1717c")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Personalized Federated Learning (PFL) aims to deliver effective client-specific models under heterogeneous distributions, yet existing methods suffer from shallow prototype alignment and brittle server-side distillation. We propose HEART-PFL, a dual-sided framework that (i) performs depth-aware Hierarchical Directional Alignment (HDA) using cosine similarity in the early stage and MSE matching in the deep stage to preserve client specificity, and (ii) stabilizes global updates through Adversarial Knowledge Transfer (AKT) with symmetric KL distillation on clean and adversarial proxy data. Using lightweight adapters with only 1.46M trainable parameters, HEART-PFL achieves stateof-the-art personalized accuracy on CIFAR-100, Flowers-102, and Caltech-101 (63.42%, 84.23%, and 95.67%, respectively) under Dirichlet non-IID partitions, and remains robust to out-of-domain proxy data. Ablation studies further confirm that HDA and AKT provide complementary gains in alignment, robustness, and optimization stability, offering insights into how the two components mutually reinforce effective personalization. Overall, these results demonstrate that HEART-PFL simultaneously enhances personalization and global stability, highlighting its potential as a strong and scalable solution for PFL (code available at https://github.com/danny0628/HEART-PFL).

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2603.24209")
get_code_for_paper("2603.24209")
have("2603.24209")

Connect an agent — have() is free.