Selected work

Representative Papers

Personalized Federated Learning for Gradient Alignment

May 03, 2026

This work addresses the challenges in personalized federated learning arising from high local gradient variance due to heterogeneous and limited client data, as well as the distortion of personalized optimization directions during model aggregation. The authors propose pFLAlign, a novel framework that, for the first time, derives a gradient alignment mechanism from a PAC-Bayesian perspective. pFLAlign employs a two-stage strategy: it adaptively adjusts gradient directions during local training and re-aligns personalized directions after global aggregation to preserve client-specific information. This approach significantly enhances both personalization performance and training stability, achieving state-of-the-art results across multiple benchmarks. Ablation studies further confirm the effectiveness of each component in the proposed framework.

0 citationsRead paper

TinyProto: Communication-Efficient Federated Learning with Sparse Prototypes in Resource-Constrained Environments

Jul 06, 2025

In resource-constrained edge federated learning, prototype-based federated learning (PFL) suffers from prohibitively high communication overhead, which scales quadratically with feature dimensionality and number of classes. To address this, we propose Class Prototype Sparsification with Adaptive Scaling (CPS-AS), a communication-efficient PFL framework that imposes structured sparsity at the class-prototype level, selectively transmits only non-zero prototype elements, and adaptively scales prototypes based on local class distributions—all without imposing additional computational burden on clients. CPS-AS thus achieves efficient, heterogeneity-aware prototype compression. Extensive experiments across multiple benchmark datasets demonstrate up to 4× reduction in communication volume, with negligible accuracy degradation (<0.5% drop), significantly outperforming existing PFL communication compression methods. The approach is particularly suitable for bandwidth-limited, device-heterogeneous edge federated deployments.

0 citationsRead paper

Heterogeneous Federated Learning with Prototype Alignment and Upscaling

Jul 06, 2025

In federated learning, data and model heterogeneity impede sufficient prototype separation, limiting the discriminative capability of prototype-based methods. To address this, we propose ProtoNorm—a server-side framework that jointly enforces prototype alignment and amplification. First, class prototypes are constrained to the unit hypersphere and optimized via a Thomson-problem-inspired formulation to promote globally uniform angular distribution, thereby enhancing inter-class separability. Second, magnitude scaling in Euclidean space is applied to further improve feature discriminability. ProtoNorm integrates prototype normalization, spherical optimization, and feature scaling without incurring additional communication overhead. Extensive experiments on multiple heterogeneous benchmark datasets demonstrate that ProtoNorm consistently outperforms state-of-the-art methods, achieving average accuracy gains of 2.1–4.7% and up to 38.5% improvement in prototype separation (e.g., inter-class cosine distance), validating its effectiveness and generalizability under resource-constrained settings.

0 citationsRead paper

Recent publications

Latest Papers

Personalized Federated Learning for Gradient Alignment

May 03, 2026

This work addresses the challenges in personalized federated learning arising from high local gradient variance due to heterogeneous and limited client data, as well as the distortion of personalized optimization directions during model aggregation. The authors propose pFLAlign, a novel framework that, for the first time, derives a gradient alignment mechanism from a PAC-Bayesian perspective. pFLAlign employs a two-stage strategy: it adaptively adjusts gradient directions during local training and re-aligns personalized directions after global aggregation to preserve client-specific information. This approach significantly enhances both personalization performance and training stability, achieving state-of-the-art results across multiple benchmarks. Ablation studies further confirm the effectiveness of each component in the proposed framework.

0 citationsRead paper

TinyProto: Communication-Efficient Federated Learning with Sparse Prototypes in Resource-Constrained Environments

Jul 06, 2025

In resource-constrained edge federated learning, prototype-based federated learning (PFL) suffers from prohibitively high communication overhead, which scales quadratically with feature dimensionality and number of classes. To address this, we propose Class Prototype Sparsification with Adaptive Scaling (CPS-AS), a communication-efficient PFL framework that imposes structured sparsity at the class-prototype level, selectively transmits only non-zero prototype elements, and adaptively scales prototypes based on local class distributions—all without imposing additional computational burden on clients. CPS-AS thus achieves efficient, heterogeneity-aware prototype compression. Extensive experiments across multiple benchmark datasets demonstrate up to 4× reduction in communication volume, with negligible accuracy degradation (<0.5% drop), significantly outperforming existing PFL communication compression methods. The approach is particularly suitable for bandwidth-limited, device-heterogeneous edge federated deployments.

0 citationsRead paper

Heterogeneous Federated Learning with Prototype Alignment and Upscaling

Jul 06, 2025

In federated learning, data and model heterogeneity impede sufficient prototype separation, limiting the discriminative capability of prototype-based methods. To address this, we propose ProtoNorm—a server-side framework that jointly enforces prototype alignment and amplification. First, class prototypes are constrained to the unit hypersphere and optimized via a Thomson-problem-inspired formulation to promote globally uniform angular distribution, thereby enhancing inter-class separability. Second, magnitude scaling in Euclidean space is applied to further improve feature discriminability. ProtoNorm integrates prototype normalization, spherical optimization, and feature scaling without incurring additional communication overhead. Extensive experiments on multiple heterogeneous benchmark datasets demonstrate that ProtoNorm consistently outperforms state-of-the-art methods, achieving average accuracy gains of 2.1–4.7% and up to 38.5% improvement in prototype separation (e.g., inter-class cosine distance), validating its effectiveness and generalizability under resource-constrained settings.

0 citationsRead paper