pFL strategies

survey — Towards Personalized Federated Learning

implementations of many

FedBN — FedBN: Federated Learning on non-IID Features via Local Batch Normalization ICLR 2021

• keep local BN layers

FedAP — FedAP: Adaptive Personalization in Federated Learning for Non-IID Data

• upon FedBN: calculate similarity pairwise matrix as weights, single model for each client in server aggregation

Regularization-based pFL

pFedMe — Personalized Federated Learning with Moreau Envelopes NeurIPS 2020

Ditto — Ditto: Fair and robust federated learning through personalization ICML 2021 code

• Each client keep a global model (for aggregation and regularization) and a local model (for personalization)
• In each round train the global model first, and train the local model with regularization similar to FedProx.

Transfer Learning

Fed-health — Fed-health: A federated transfer learning framework for wearable healthcare

Meta-learning

Per-FedAvg (MAML) — Personalized Federated Learning with Theoretical Guarantees: A Model-Agnostic Meta-Learning Approach NeurIPS 2020

Multi-Task Learning

MOCHA — Federated Multi-Task Learning

User clustering

Three Approaches for Personalization with Applications to Federated Learning | code

ClusterFL — ClusterFL: A Similarity-Aware Federated Learning System for Human Activity Recognition

ClusteredFL — Clustered federated learning: Model- agnostic distributed multitask optimization under privacy constraints

Personalized aggregation

APFL — Adaptive Personalized Federated Learning

• three models for each client: global, local and personalized mixture of global and local models with an $\alpha$
• adaptively update the alpha using gradient descent | code

FedALA — FedALA: Adaptive Local Aggregation for Personalized Federated Learning | code

• adaptively learning element-wise aggregation weights

FedFomo — Personalized Federated Learning with First Order Model Optimization ICLR 2021

• client download multiple models from server and approximate weights for aggregation

FedPHP — FedPHP: Federated Personalization with Inherited Private Models ECML PKDD 2021

• use temporal ensembling (?) of a client' historical personalized models to supervise the personalization process in the next global round

FedAMP — Personalized Cross-Silo Federated Learning on non-IID Data AAAI 2021

APPLE — Adapt to Adaptation: Learning Personalization for Cross-Silo Federated Learning IJCAI 2022

Feature (representation)

FedPer — Federated Learning with Personalization Layers

• base + personalization layer

FedRep — Exploiting Shared Representations for Personalized Federated Learning ICML 2021

• shared data representation across clients and unique local heads for each client