Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
FedL2P: Federated Learning to Personalize
Authors: Royson Lee, Minyoung Kim, Da Li, Xinchi Qiu, Timothy Hospedales, Ferenc Huszar, Nicholas Lane
NeurIPS 2023 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results show that this framework improves on a range of standard hand-crafted personalization baselines in both label and feature shift situations.0 |
| Researcher Affiliation | Collaboration | 1 University of Cambridge, UK 2 Samsung AI Center, Cambridge, UK 3 University of Edinburgh, UK 4 Flower Labs |
| Pseudocode | Yes | Algorithm 1 Fed L2P: FL of meta-nets for Personalization Hyperparameters; Algorithm 2 Hypergradient |
| Open Source Code | Yes | Code is available at https://github.com/royson/fedl2p |
| Open Datasets | Yes | CIFAR10 [31]. A widely-used image classification dataset, also popular as an FL benchmark. The number of clients C is set to 1000 and 20% of the training data is used for validation. |
| Dataset Splits | Yes | CIFAR10 [31]. A widely-used image classification dataset, also popular as an FL benchmark. The number of clients C is set to 1000 and 20% of the training data is used for validation. |
| Hardware Specification | Yes | We use the Flower federated learning framework [6] and 8 NVIDIA Ge Force RTX 2080 Ti GPUs for all experiments. |
| Software Dependencies | No | The paper mentions 'Flower federated learning framework' and 'torchvision [41]' but does not provide specific version numbers for these or other software dependencies. |
| Experiment Setup | Yes | The learning rate (ζ) for λ = {wbn, wlr, η} is set to {10 3,10 3,10 4}, respectively. The hypergradient is clipped by value [ 1, 1], Q = 3, and ψ = 0.1 in Alg. 2. The maximum number of communication rounds is set to 500, and over the rounds we save the λ value that leads to the lowest validation loss, averaged over the participating clients, as the final learned λ. The fraction ratio r=0.1 unless stated otherwise, sampling 10% of the total number of clients per FL round. |