Janus: Dual-Server Multi-Round Secure Aggregation with Verifiability for Federated Learning
Authors: Lang Pu, Jingjing Gu, Chao Lin, Xinyi Huang
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To assess the practical performance of Janus, we carried out comprehensive experiments focusing on both effectiveness and efficiency. We further compared it with several representative advanced schemes. Our experimental setup includes a 13th Gen Intel(R) Core(TM) i7-13700KF 3.40 GHz processor with 32.0 GB of RAM, a 64-bit Windows 11 operating system, and an RTX 4070Ti GPU display adapter. |
| Researcher Affiliation | Academia | 1College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing, China 2College of Computer and Cyber Security, Fujian Normal University, Fuzhou, China 3College of Cyber Security, Jinan University, Guangzhou, China. Correspondence to: Jingjing Gu < EMAIL>, Chao Lin < EMAIL>. |
| Pseudocode | Yes | Figure 4. Detailed Construction of Janus. All parties get the security parameter λ. This phase generates the public parameter pp of the system, which contains the specific commitment, one-time pad, and public key encryption. The assitant server S1 generates its public/private keys (pks, sks) and publishes its public key to all users. Each user generates its public/private keys (pki, ski) and publish its public key to servers S0 and S1. Subsequent user-server interactions are via public key encryption by default. |
| Open Source Code | No | No explicit statement about providing open-source code for the described methodology or a link to a repository is provided in the paper. |
| Open Datasets | Yes | Datasets and Models. MNIST consists of 70,000 grayscale handwritten digit images (60,000 for training, 10,000 for testing), each 28x28 pixels. ... CIFAR-10 includes 60,000 color images across 10 classes (50,000 training, 10,000 testing)... |
| Dataset Splits | Yes | MNIST consists of 70,000 grayscale handwritten digit images (60,000 for training, 10,000 for testing), each 28x28 pixels. ... CIFAR-10 includes 60,000 color images across 10 classes (50,000 training, 10,000 testing)... |
| Hardware Specification | Yes | Our experimental setup includes a 13th Gen Intel(R) Core(TM) i7-13700KF 3.40 GHz processor with 32.0 GB of RAM, a 64-bit Windows 11 operating system, and an RTX 4070Ti GPU display adapter. |
| Software Dependencies | No | The paper mentions 'a 64-bit Windows 11 operating system' but does not specify any other software dependencies like programming languages, libraries, or frameworks with version numbers. |
| Experiment Setup | Yes | We simulated an environment with 100 users, each holding 600 local training samples. The global model for MNIST is a fully connected network with layers of size (784, 256, 10). CIFAR-10 includes 60,000 color images across 10 classes (50,000 training, 10,000 testing), using a CNN architecture with a batch size of 10, a learning rate of 0.001, and 100 training epochs. We employed SGD as the optimizer, with each user applying SGD once per global epoch. |