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..
Federated Learning via Meta-Variational Dropout
Authors: Insu Jeon, Minui Hong, Junhyeog Yun, Gunhee Kim
NeurIPS 2023 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conducted extensive experiments on various sparse and non-IID FL datasets. Meta VD demonstrated excellent classification accuracy and uncertainty calibration performance, especially for out-of-distribution (OOD) clients. |
| Researcher Affiliation | Academia | Insu Jeon Minui Hong Junhyeog Yun Gunhee Kim Seoul National University, Seoul, South Korea EMAIL EMAIL |
| Pseudocode | Yes | Algorithm 1 Meta VD algorithm with MAML and Reptile variant for FL |
| Open Source Code | Yes | Code is available at https://github.com/insujeon/Meta VD. ... Our code is available at https://github.com/insujeon/Meta VD. |
| Open Datasets | Yes | We used multiple FL datasets [72], including CIFAR-10, CIFAR-100, FEMINIST, and Celeb A. ... The CIFAR-10 and CIFAR-100 datasets [98] are popular for 10-class and 100-class image classification respectively. ... The Federated Extended MNIST (FEMNIST) is a widely used FL dataset for 62-class handwritten character recognition [72]. ... The Celeb A is a FL dataset based on [100] for 2-class image classification. |
| Dataset Splits | Yes | procedure LOCALADAPTATION_MAML(θ, α) ... Sample dataset Dm tr and Dm val from Dm |
| Hardware Specification | Yes | All experiments are run on a cluster of 32 NVIDIA GTX 1080 GPUs. |
| Software Dependencies | No | The paper mentions 'an optimization tool called Optuna[103]' but does not provide specific version numbers for general software dependencies like Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | The batch size was set to 64, and local steps was set to 5. Personalization was executed with a batch size of 64 and a 1-step update. ... The server learning rate η was explored within the range of [0.6, 0.7, 0.8, 0.9, 1.0]. The local SGD learning rate was investigated within the range of [0.005, 0.01, 0.015, 0.02, 0.025, 0.03]. ... For Meta VD, an additional KL divergence parameter β is needed, and we sought its optimal value within the range of [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]. |