Fair and Efficient Contribution Valuation for Vertical Federated Learning
Authors: Zhenan Fan, Huang Fang, Xinglu Wang, Zirui Zhou, Jian Pei, Michael Friedlander, Yong Zhang
ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Both theoretical analysis and extensive experimental results demonstrate the fairness, efficiency, adaptability, and effectiveness of Ver Fed SV. ... 7 EXPERIMENTS We conduct extensive experiments on real-world datasets, including Adult (Zeng et al., 2008), Web (Platt, 1998), Covtype (Blackard & Dean, 1999), and RCV1 (Lewis et al., 2004). |
| Researcher Affiliation | Collaboration | University of British Columbia, {zhenanf,hgfang, mpf}@cs.ubc.ca Simon Fraser University, {xinglu wang 2,jpei}@cs.sfu.ca Huawei Technologies Canada, {zirui.zhou, yong.zhang3}@huawei.com |
| Pseudocode | No | The paper describes algorithms (e.g., Fed SGD in Section A.1, VAFL in Section A.2) in prose, but does not present them as structured pseudocode or algorithm blocks with formal labels. |
| Open Source Code | Yes | Our code is submitted in the supplementary material. |
| Open Datasets | Yes | We use four real-world data sets, Adult (Zeng et al., 2008), Web (Platt, 1998), Covtype (Blackard & Dean, 1999), and RCV1 (Lewis et al., 2004) 1. ... 1From the website of LIBSVM (Chang & Lin, 2011) https://www.csie.ntu.edu.tw/ cjlin/ libsvmtools/datasets/ |
| Dataset Splits | No | The paper mentions "training", "validation", and "test" but does not specify how the datasets are split into these sets (e.g., percentages, sample counts, or specific methods for partitioning). |
| Hardware Specification | Yes | All the experiments are conducted on a Linux server with 32 CPUs and 64 GB memory. |
| Software Dependencies | No | The paper mentions the software used: "We implement the VFL algorithms and the corresponding Ver Fed SV computation schemes described in Sections 5 and 6 in the Julia language (Bezanson et al., 2017). ... The matrix completion problem in Equation 2 is solved by the Julia package Low Rank Models.jl (Udell et al., 2016)." However, it does not specify version numbers for Julia or the Low Rank Models.jl package. |
| Experiment Setup | Yes | Hyperparameter setting (synchronous) We summarize the hyperparameters under the synchronous setting in Table 3, where η and τ are the learning rate and the batch size used in the Fed SGD algorithm, and r and λ are the rank parameter and regularization parameter used in the matrix completion problem. ... Hyperparameter setting (asynchronous) We summarize the hyperparameters under the asynchronous setting in Table 4, where η and τ are the learning rate and the batch size used in the VAFL algorithm, and t is the communication frequency for clients. |