Scalable Federated Unlearning via Isolated and Coded Sharding
Authors: Yijing Lin, Zhipeng Gao, Hongyang Du, Dusit Niyato, Gui Gui, Shuguang Cui, Jinke Ren
IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, extensive experiments on two typical learning tasks, i.e., classification and generation, demonstrate that our proposed framework can achieve better performance than three state-of-the-art frameworks in terms of accuracy, retraining time, storage overhead, and F1 scores for resisting membership inference attacks. |
| Researcher Affiliation | Academia | Yijing Lin1,2 , Zhipeng Gao1 , Hongyang Du4 , Dusit Niyato4 , Gui Gui5 , Shuguang Cui3,2 , Jinke Ren2,3 1 State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications 2 The Future Network of Intelligence Institute, The Chinese University of Hong Kong (Shenzhen) 3 School of Science and Engineering, The Chinese University of Hong Kong (Shenzhen) 4 College of Computing and Data Science, Nanyang Technological University 5 School of Automation, Central South University |
| Pseudocode | Yes | Algorithm 1 Federated Unlearning with Isolated Sharding and Coded Computing |
| Open Source Code | No | The paper refers to 'Nano GPT1 [Radford et al., 2019]' with a footnote '1https://github.com/karpathy/nano GPT'. This link is for a third-party model used in the experiments, not for the authors' own implementation of their proposed framework. No explicit statement about releasing their own code was found. |
| Open Datasets | Yes | We use four commonly-adopted datasets including MNIST [Le Cun et al., 1998], Fashion MNIST [Xiao et al., 2017], CIFAR-10 [Krizhevsky et al., 2009], and Tiny Shakespeare [Mc Mahan et al., 2017] for experiments, which are applicable to a diverse range of tasks. |
| Dataset Splits | No | The paper states: 'In the learning process, only 20 clients are randomly selected in each training round. These clients are divided into 4 shards such that each shard has 5 clients. The numbers of local epochs and training rounds are set as 10 and 30, respectively.' While it describes the training process, it does not explicitly provide information on validation dataset splits (e.g., percentages, counts, or specific methods like cross-validation). |
| Hardware Specification | No | The paper does not provide specific details regarding the hardware specifications (e.g., GPU/CPU models, memory, or cloud resources) used for conducting the experiments. |
| Software Dependencies | No | The paper mentions 'Nano GPT' and provides a GitHub link, but it does not specify any general software dependencies or libraries with their version numbers (e.g., Python, TensorFlow, PyTorch versions) used for the implementation of their framework or experiments. |
| Experiment Setup | Yes | We consider a total of 100 clients in our experiments to demonstrate the effectiveness of the proposed framework. In the learning process, only 20 clients are randomly selected in each training round. These clients are divided into 4 shards such that each shard has 5 clients. The numbers of local epochs and training rounds are set as 10 and 30, respectively. |