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 [1].
Dyn-D^2P: Dynamic Differentially Private Decentralized Learning with Provable Utility Guarantee
Authors: Zehan Zhu, Yan Huang, Xin Wang, Shouling Ji, Jinming Xu
IJCAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on benchmark datasets demonstrate the superiority of Dyn-D2P over its counterparts employing fixed-level noises, especially under strong privacy guarantees. Furthermore, we provide a provable utility bound for Dyn-D2P that establishes an explicit dependency on network-related parameters, with a scaling factor of 1/ n in terms of the number of nodes n up to a bias error term induced by gradient clipping. |
| Researcher Affiliation | Academia | 1Zhejiang University, Hangzhou, China 2Qilu University of Technology, Jinan, China EMAIL, EMAIL, EMAIL |
| Pseudocode | Yes | The complete pseudocode is summarized in Algorithm 1. At a high level, Dyn-D2P is comprised of local SGD and the averaging of neighboring information, following a framework similar to SGP [Assran et al., 2019] which employs the Push Sum protocol [Kempe et al., 2003] to tackle the unblanceness of directed graphs. However, the key distinction lies in the gradient clipping operation and the injection of DP Gaussian Algorithm 1 Dyn-D2P |
| Open Source Code | No | The paper does not provide explicit statements about the release of source code for the methodology described, nor does it provide a direct link to a code repository. It mentions pseudocode can be found in the full version/appendix, but this is not the same as providing source code. |
| Open Datasets | Yes | We compare five algorithms in a fully decentralized setting composed of 20 nodes, on two benchmark non-convex learning tasks: i) training Res Net-18 [He et al., 2016] on Cifar-10 [Krizhevsky, 2009] dataset; ii) training shallow CNN model (composed of two convolution layers and two fully connected layers) on Fashion Mnist [Xiao et al., 2017] dataset. |
| Dataset Splits | Yes | We split shuffled datasets evenly to 20 nodes. For communication topology, unless otherwise stated, we use a time-varying directed exponential graph (refer to Appendix E in our full version [Zhu et al., 2025] for its definition). |
| Hardware Specification | Yes | All experiments are deployed in a server with Intel Xeon E5-2680 v4 CPU @ 2.40GHz and 8 Nvidia RTX 3090 GPUs, and are implemented with distributed communication package torch.distributed in Py Torch [Paszke et al., 2017], where a process serves as a node, and inter-process communication is used to mimic communication among nodes. |
| Software Dependencies | No | The paper mentions being "implemented with distributed communication package torch.distributed in Py Torch [Paszke et al., 2017]" but does not specify a version number for PyTorch or any other software dependencies, which is required for reproducibility. |
| Experiment Setup | Yes | The learning rate is set to be 0.05 for Res Net-18 training and 0.03 for shallow CNN model training. Privacy parameters ฮด is set to be 10 4, and we test different values for ฯต which implies different levels of privacy guarantee. Other parameters such as C, C0, ฯc and ฯยต are detailed in Appendix F.1 in our full version [Zhu et al., 2025]. |