Federated Representation Learning in the Under-Parameterized Regime
Authors: Renpu Liu, Cong Shen, Jing Yang
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate that FLUTE outperforms state-of-the-art FRL solutions in both synthetic and real-world tasks. and 7. Experimental Results |
| Researcher Affiliation | Academia | 1Department of Electrical Engineering, The Pennsylvania State University, University Park, PA, USA 2Department of Electrical and Computer Engineering, University of Virginia, Charlottesville, Virginia, USA. |
| Pseudocode | Yes | Algorithm 1 FLUTE Linear and Algorithm 2 in Appendix B. |
| Open Source Code | Yes | Main experiments can be reproduced with the code provided under the following link: https://github.com/ Renpu Liu/flute |
| Open Datasets | Yes | We conduct a series of experiments utilizing both synthetic datasets for linear FLUTE and real-world datasets, specifically CIFAR-10 and CIFAR-100 (Krizhevsky et al., 2009), for general FLUTE. |
| Dataset Splits | No | The paper mentions 'N' samples per client and 'm' classes per client, and evaluates 'average test accuracy' but does not specify the explicit train/validation/test dataset splits with percentages or counts. |
| Hardware Specification | No | The paper discusses hardware limitations for federated learning (e.g., 'Raspberry Pi 4', 'IoT devices') as motivation but does not specify the hardware used to run the experiments described in the paper. |
| Software Dependencies | No | The paper describes using convolutional neural networks but does not specify the software libraries or their version numbers used for implementation (e.g., TensorFlow, PyTorch versions). |
| Experiment Setup | Yes | Input: Learning rates ηl and ηr, regularization parameter λ, communication round T, constant α, The learning rate is set to η = 0.03, and for random initialization, we set α = 1 10d., all algorithms are executed over 100 communication rounds., The number of local updates for LG-Fed, Fed Per, Fed Ro D, Fed CP and FLUTE are set to 5. |