Secure Distributed Sparse Gaussian Process Models Using Multi-Key Homomorphic Encryption
Authors: Adil Nawaz, Guopeng Chen, Muhammad Umair Raza, Zahid Iqbal, Jianqiang Li, Victor C.M. Leung, Jie Chen
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show that with a little sacrifice in terms of time complexity, we achieve a secure d GP model without deteriorating the predictive performance compared to traditional nonsecure d GP models. We also present a practical implementation of the proposed model using 15 Nvidia Jetson Nano Developer Kit modules to simulate a real-world scenario. |
| Researcher Affiliation | Academia | 1College of Computer Science & Software Engineering, Shenzhen University, Shenzhen, China 2 College of Electronics & Information Engineering, Shenzhen University, Shenzhen, China 3 National Engineering Laboratory for Big Data System Computing Technology, Shenzhen University, Shenzhen, China |
| Pseudocode | No | The paper includes a schematic diagram (Figure 1) which illustrates the framework stages, but no structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper states 'The source code for the framework is developed in C++ with HE implementation based on HEAAN V2.1.0.' but does not provide any link or explicit statement about its public availability. |
| Open Datasets | Yes | The proposed scheme is evaluated on the three publicly available regression datasets, i.e, Sarcos (Vijayakumar and Schaal 2000), Power Consumption (Pow Cons) (Salam and Hibaoui 2018) and SGEMM GPU kernel performance (Nugteren and Codreanu 2015) dataset. |
| Dataset Splits | No | In all the experiments, after standardizing the datasets, we split them into training and test sets, where test set in each case comprises of 10% of the whole dataset. A separate validation split is not mentioned. |
| Hardware Specification | Yes | All experiments are implemented using 15 Nvidia Jetson Nano 4GB developer kit modules as the client nodes and an Intel(R) Core(TM) i5-5500U CPU @ 2.40GHz with 8GB RAM as a server connected via a 10/100Mbps standard Ethernet WLAN network. |
| Software Dependencies | Yes | The source code for the framework is developed in C++ with HE implementation based on HEAAN V2.1.0. HE operations involve computation on big numbers and modulo arithmetic which is implemented using NTL 11.5.1 built in conjunction with GMP 6.2.1. |
| Experiment Setup | Yes | The parameters for HE, N = 216, q = 800, h = 64, ρ = 0.5 and σ = 3.2. In all the experiments, after standardizing the datasets, we split them into training and test sets, where test set in each case comprises of 10% of the whole dataset. The datasets are modelled as GP specified by the squared exponential kernel k(u, u ) such that, k(u, u ) σ2 a exp( 1 2 Pd i=1( ui u i ℓi 2 ) + σ2 pδuu , where ui, u i denote the i-th feature(s) of d-dimensional input vector(s) u, u respectively. σa, σp and li are the hyperparameters representing function variance, noise variance and length-scales respectively. δuu is the Kronecker delta which is 1 if u = u and 0 otherwise. Hyperparameter values are learned by Maximum Likelihood Estimation (MLE) (Rasmussen 2003). |