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; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]

Budgeted Online Model Selection and Fine-Tuning via Federated Learning

Authors: Pouya M. Ghari, Yanning Shen

TMLR 2024 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on real datasets demonstrate the effectiveness of the proposed algorithm. Experiments on regression and image classification datasets showcase the effectiveness of OFMS-FT compared with state-of-the-art alternatives.
Researcher Affiliation Academia Pouya M. Ghari EMAIL Department of Electrical Engineering and Computer Science Center for Pervasive Communications and Computing University of California Irvine; Yanning Shen EMAIL Department of Electrical Engineering and Computer Science Center for Pervasive Communications and Computing University of California Irvine
Pseudocode Yes Algorithm 1 OFMS-FT: Online Federated Model Selection and Fine-Tuning; Algorithm 2 Cluster Generation by Client i at Learning Round t
Open Source Code No The paper does not contain any explicit statements about releasing source code, nor does it provide links to a code repository.
Open Datasets Yes Image classification tasks are performed over CIFAR-10 (Krizhevsky, 2009) and MNIST (Lecun et al., 1998) datasets. Online regression tasks are performed on Air (Zhang et al., 2017) and WEC (Neshat et al., 2018) datasets.
Dataset Splits Yes For CIFAR-10, each client receives 155 testing data samples from one class and 5 samples from each of the other nine classes. In the case of MNIST, each client receives at least 133 samples from one class and at least 5 samples from the other classes. The 200 testing data samples are randomly shuffled and are sequentially presented to each client over T = 200 learning rounds.
Hardware Specification Yes All experiments were carried out using Intel(R) Core(TM) i7-10510U CPU @ 1.80 GHz 2.30 GHz processor with a 64-bit Windows operating system.
Software Dependencies No The paper mentions software components like ReLU activation, SGD, and VGG architecture, but it does not provide specific version numbers for any libraries or frameworks used in the implementation.
Experiment Setup Yes The learning rates ηi for all methods are set to be 10/T where T = 200. Furthermore, the fine-tuning learning rate is set to ηf = 10^-3 / T. In the experiments, using Fed-OMD and Per Fed Avg, each client performs one epoch of stochastic gradient descent (SGD) with learning rate of 0.001 on its batch of data to fine-tune the model.