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 [1].
Fast Federated Learning in the Presence of Arbitrary Device Unavailability
Authors: Xinran Gu, Kaixuan Huang, Jingzhao Zhang, Longbo Huang
NeurIPS 2021 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We also provide an explicit characterization of the improvement over baseline algorithms through a case study, and validate the results by numerical experiments on real-world datasets.In this section, we conduct numerical experiments3. to verify our theoretical results and investigate how the heterogeneity of the device availability influences the Federated optimization algorithms. |
| Researcher Affiliation | Academia | Xinran Gu IIIS Tsinghua University EMAIL Kaixuan Huang ECE Princeton University EMAIL Jingzhao Zhang EECS Massachusetts Institute of Technology EMAIL Longbo Huang IIIS Tsinghua University EMAIL |
| Pseudocode | Yes | Algorithm 1 Memory-augmented Impatient Federated Averaging (MIFA) |
| Open Source Code | Yes | Our code is available at https://github.com/hmgxr128/MIFA_code/ |
| Open Datasets | Yes | Following [26, 25], we construct non-i.i.d. datasets from two commonly used computer vision datasets MNIST [23] and CIFAR-10 [22]. |
| Dataset Splits | No | The paper uses commonly used datasets (MNIST, CIFAR-10) but does not explicitly detail training, validation, or test dataset splits. |
| Hardware Specification | Yes | We run all the experiments with 4 GPUs of type Ge Force RTX 2080 Ti. |
| Software Dependencies | No | The paper mentions adapting code from a previous work [26] but does not provide specific version numbers for software dependencies or libraries used for the experiments. |
| Experiment Setup | Yes | In all the experiments, we set the initial learning rate to be η0 = 0.1 and decay the learning rate as ηt = η0 1 t . We set the weight decay to be 0.001. The local batch size is 100 and each local update consists of 2 epochs. |