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..
Resource-Constrained Federated Continual Learning: What Does Matter?
Authors: Yichen Li, Yuying Wang, Jiahua Dong, Haozhao Wang, Yining Qi, Rui Zhang, Ruixuan Li
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Through extensive experiments amounting to a total of over 1,000+ GPU hours, we find that, under limited resource-constrained settings, existing FCL approaches, with no exception, fail to achieve the expected performance. Our conclusions are consistent in the sensitivity analysis. |
| Researcher Affiliation | Academia | 1School of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan, China 2Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi, United Arab Emirates 3School of Computer Science and Technology, Soochow University, Suzhou, China |
| Pseudocode | No | The paper describes methods verbally and provides problem formulations in Appendix B using mathematical equations, but it does not contain any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | In the NeurIPS Paper Checklist, under '5. Open access to data and code', the justification states: 'We could provide our code if required.' This is a conditional statement and does not constitute concrete access. |
| Open Datasets | Yes | We conduct our experiments with heterogeneous datasets over two typical scenarios: Class Incremental Learning and Domain-Incremental Learning on six datasets: CIFAR-10 [18], CIFAR-100 [18], Tiny-Image Net [20], Digit-10, Office-31 [50] and Office-Caltech-10 [69]. |
| Dataset Splits | Yes | CIFAR-10 [18]: Comprises 10 object classes... It includes 50,000 training and 10,000 test images. CIFAR-100 [18]: ... It contains 50,000 training and 10,000 test images. Tiny-Image Net [20]: ... It includes 100,000 training images, 10,000 validation images, and 10,000 test images. |
| Hardware Specification | Yes | All experiments are run on 8 RTX 4090 GPUs and 8*2 RTX 3090 GPUs. |
| Software Dependencies | No | The paper mentions using 'Adam as an optimizer' and refers to parameters from 'original open-source code' but does not specify version numbers for any software, libraries, or frameworks used in their implementation. |
| Experiment Setup | Yes | We use Adam as an optimizer with a linear learning rate schedule. We set the remaining parameters according to the values in the original open-source code, such as the weight of multiple distillation losses. Table 5 provides specific details for each dataset including Batch Size, Learning Rate, Local training epoch, Client selection ratio, and Communication Round. |