Secure Bilevel Asynchronous Vertical Federated Learning with Backward Updating
Authors: Qingsong Zhang, Bin Gu, Cheng Deng, Heng Huang10896-10904
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on benchmark datasets demonstrate that our algorithms are efficient, scalable and lossless. In this section, extensive experiments are conducted to demonstrate the efficiency, scalability and losslessness of our algorithms. |
| Researcher Affiliation | Collaboration | 1School of Electronic Engineering, Xidian University, Xi an 710071, China; 2JD Tech, Beijing 100176, China. 3JD Finance America Corporation, Mountain View, CA, USA; 4MBZUAI, United Arab Emirates 5Electrical and Computer Engineering, University of Pittsburgh, PA, USA |
| Pseudocode | Yes | Algorithm 1 Safe algorithm of obtaining w T xi. Algorithm 2 VFB2-SGD for active party ℓto actively launch dominated updates. Algorithm 3 VFB2-SGD for the ℓ-th party to passively launch collaborative updates. |
| Open Source Code | No | The paper does not provide any specific links or explicit statements about the availability of source code for the described methodology. |
| Open Datasets | Yes | Datasets: We use four classification datasets summarized in Table 1 for evaluation. Especially, D1 (UCICredit Card) and D2 (Give Me Some Credit) are the real financial datsets from the Kaggle website , which can be used to demonstrate the ability to address real-world tasks; D3 (news20) and D4 (webspam) are the large-scale ones from the LIBSVM (Chang and Lin 2011) website . |
| Dataset Splits | No | The paper states: 'We use the training dataset or randomly select 80% samples as the training data, and the testing dataset or the rest as the testing data.' This describes a train/test split but does not explicitly mention a separate validation set or provide details for creating one. |
| Hardware Specification | No | The paper states: 'All experiments are implemented on a machine with four sockets, and each sockets has 12 cores.' This describes the core count but does not provide specific hardware models (e.g., CPU, GPU) or memory details for reproducibility. |
| Software Dependencies | No | The paper mentions 'We use MPI to implement the communication scheme.' However, it does not provide specific version numbers for MPI or any other software dependencies, which is required for reproducibility. |
| Experiment Setup | Yes | An optimal learning rate γ is chosen from {5e 1, 1e 1, 5e 2, 1e 2, } with regularization coefficient λ = 1e 4 for all experiments. We set q = 8, m = 3 and fix the γ for algorithms with a same SGD-type but in different parallel fashions. Each comparison is repeated 10 times with m = 3, q = 8, and a same stop criterion, e.g., 1e 5 for D1. |