4,996 Hits in 3.8 sec

Network Generation with Differential Privacy [article]

Xu Zheng, Nicholas McCarthy, Jer Hayes
2021 arXiv   pre-print
Differential privacy is a gold standard for data privacy, and the introduction of the differentially private stochastic gradient descent (DP-SGD) algorithm has facilitated the training of private neural  ...  We consider the problem of generating private synthetic versions of real-world graphs containing private information while maintaining the utility of generated graphs.  ...  The main idea behind these approaches is to model the probabilities of each edge's existence by a neural network.  ... 
arXiv:2111.09085v1 fatcat:4gtssq7k6nbb5do74dlmvjlnse

Locally Private Graph Neural Networks [article]

Sina Sajadmanesh, Daniel Gatica-Perez
2021 arXiv   pre-print
Graph Neural Networks (GNNs) have demonstrated superior performance in learning node representations for various graph inference tasks.  ...  In this paper, we study the problem of node data privacy, where graph nodes have potentially sensitive data that is kept private, but they could be beneficial for a central server for training a GNN over  ...  RELATED WORK Graph neural networks.  ... 
arXiv:2006.05535v9 fatcat:xqzrddss6zhzhgzcllxvq53qeu

Locally Private Graph Neural Networks

Sina Sajadmanesh, Daniel Gatica-Perez
2021 Zenodo  
Graph Neural Networks (GNNs) have demonstrated superior performance in learning node representations for various graph inference tasks.  ...  In this paper, we study the problem of node data privacy, where graph nodes have potentially sensitive data that is kept private, but they could be beneficial for a central server for training a GNN over  ...  RELATED WORK Graph neural networks.  ... 
doi:10.5281/zenodo.5081878 fatcat:5s4qnep7trbi3pjvmpz5fbg3kq

Releasing Graph Neural Networks with Differential Privacy Guarantees [article]

Iyiola E. Olatunji, Thorben Funke, Megha Khosla
2021 arXiv   pre-print
With the increasing popularity of Graph Neural Networks (GNNs) in several sensitive applications like healthcare and medicine, concerns have been raised over the privacy aspects of trained GNNs.  ...  The student GNN is trained using public data, partly labeled privately using the teacher GNN models trained exclusively for each query node.  ...  RELATED WORKS Graph neural networks (GNNs) [8, 12, 23] mainly popularized by graph convolution networks and their variants compute node representations by recursive aggregation and transformation of  ... 
arXiv:2109.08907v1 fatcat:gl6tfcsn5fgmzdvg5nnrx377aq

Secure Deep Graph Generation with Link Differential Privacy [article]

Carl Yang, Haonan Wang, Ke Zhang, Liang Chen, Lichao Sun
2021 arXiv   pre-print
Many data mining and analytical tasks rely on the abstraction of networks (graphs) to summarize relational structures among individuals (nodes).  ...  In this paper, we leverage the differential privacy (DP) framework to formulate and enforce rigorous privacy constraints on deep graph generation models, with a focus on edge-DP to guarantee individual  ...  The randomized mechanism M f (G) is (ε, δ)-differentially private if σ ≥ 2 ln(1.25/δ)/ε and ε < 1. In our setting, G is the original training graph.  ... 
arXiv:2005.00455v3 fatcat:uwlt42ndqrctlbuek3w527guze

Sensitivity analysis in differentially private machine learning using hybrid automatic differentiation [article]

Alexander Ziller, Dmitrii Usynin, Moritz Knolle, Kritika Prakash, Andrew Trask, Rickmer Braren, Marcus Makowski, Daniel Rueckert, Georgios Kaissis
2021 arXiv   pre-print
This enables modelling the sensitivity of arbitrary differentiable function compositions, such as the training of neural networks on private data.  ...  Moreover, we investigate the application of our technique to the training of DP neural networks.  ...  On a high level, AD stores computations in a graph data structure in which every node maintains a reference to its parents as well as the operation from which it originated.  ... 
arXiv:2107.04265v2 fatcat:jdu45azlafbe5bombwxkqeorga

Differentially Private Language Models Benefit from Public Pre-training [article]

Gavin Kerrigan and Dylan Slack and Jens Tuyls
2020 arXiv   pre-print
When training a language model on sensitive information, differential privacy (DP) allows us to quantify the degree to which our private data is protected.  ...  However, training algorithms which enforce differential privacy often lead to degradation in model quality.  ...  Additionally, there is not a discernible difference between the various levels of private fine-tuning. This is likely because feedforward neural networks are not strong language models.  ... 
arXiv:2009.05886v2 fatcat:gqoxj62nqjdbtjxaoc7wof3dyi

Examining COVID-19 Forecasting using Spatio-Temporal Graph Neural Networks [article]

Amol Kapoor, Xue Ben, Luyang Liu, Bryan Perozzi, Matt Barnes, Martin Blais, Shawn O'Banion
2020 arXiv   pre-print
We evaluate this approach on the US county level COVID-19 dataset, and demonstrate that the rich spatial and temporal information leveraged by the graph neural network allows the model to learn complex  ...  In this work, we examine a novel forecasting approach for COVID-19 case prediction that uses Graph Neural Networks and mobility data.  ...  METHOD 3.1 Graph Neural Networks The core insight behind graph neural network models is that the transformation of the input node's signal can be coupled with the propagation of information from a node's  ... 
arXiv:2007.03113v1 fatcat:z5qtqfezovcxhgzvjyfgz6hexq

PyTorch Geometric Temporal: Spatiotemporal Signal Processing with Neural Machine Learning Models [article]

Benedek Rozemberczki and Paul Scherer and Yixuan He and George Panagopoulos and Alexander Riedel and Maria Astefanoaei and Oliver Kiss and Ferenc Beres and Guzmán López and Nicolas Collignon and Rik Sarkar
2021 arXiv   pre-print
PyTorch Geometric Temporal was created with foundations on existing libraries in the PyTorch eco-system, streamlined neural network layer definitions, temporal snapshot generators for batching, and integrated  ...  We present PyTorch Geometric Temporal a deep learning framework combining state-of-the-art machine learning algorithms for neural spatiotemporal signal processing.  ...  The framework reuses the existing high level neural network layer classes as building blocks from the PyTorch and PyTorch Geometric ecosystems.  ... 
arXiv:2104.07788v3 fatcat:ktnji6kjrzd7no6blp6zimfutu

Privacy-Preserving Graph Convolutional Networks for Text Classification [article]

Timour Igamberdiev, Ivan Habernal
2021 arXiv   pre-print
Graph convolutional networks (GCNs) are a powerful architecture for representation learning on documents that naturally occur as graphs, e.g., citation or social networks.  ...  We address these challenges by adapting differentially-private gradient-based training to GCNs and conduct experiments using two optimizers on five NLP datasets in two languages.  ...  To the best of our knowledge, this is the first study that brings differentially private gradient-based training to graph neural networks.  ... 
arXiv:2102.09604v2 fatcat:mkypqtsx6rc6hdoflcd7fw4v2e

FedGNN: Federated Graph Neural Network for Privacy-Preserving Recommendation [article]

Chuhan Wu, Fangzhao Wu, Yang Cao, Yongfeng Huang, Xing Xie
2021 arXiv   pre-print
Graph neural network (GNN) is widely used for recommendation to model high-order interactions between users and items.  ...  Since local gradients may contain private information, we apply local differential privacy techniques to the local gradients to protect user privacy.  ...  Various kinds of GNN network can be used in our framework, such as graph convolution network (GCN) [13] , gated graph neural network (GGNN) [16] and graph attention network (GAT) [28] .  ... 
arXiv:2102.04925v2 fatcat:cucnmoawcfhtvkz4gtcd6ovdoi

ReGVD: Revisiting Graph Neural Networks for Vulnerability Detection [article]

Van-Anh Nguyen and Dai Quoc Nguyen and Van Nguyen and Trung Le and Quan Hung Tran and Dinh Phung
2021 arXiv   pre-print
Inspired by the successful applications of pre-trained programming language (PL) models such as CodeBERT and graph neural networks (GNNs), we propose ReGVD, a general and novel graph neural network-based  ...  Next, ReGVD leverages a practical advantage of residual connection among GNN layers and explores a beneficial mixture of graph-level sum and max poolings to return a graph embedding for the given source  ...  Graph neural networks (GNNs) have recently become a central method to embed nodes and graphs into low-dimensional continuous vector spaces (Hamilton et al., 2017; Wu et al., 2019) .  ... 
arXiv:2110.07317v1 fatcat:bsor5auiq5g4xbvguw3pwckxdq

Privacy-Preserving Representation Learning on Graphs: A Mutual Information Perspective [article]

Binghui Wang, Jiayi Guo, Ang Li, Yiran Chen, Hai Li
2021 arXiv   pre-print
Next, we train these parameterized neural networks to approximate the true mutual information and learn privacy-preserving node representations.  ...  Then, we derive tractable variational bounds for the mutual information terms, where each bound can be parameterized via a neural network.  ...  Similarly, we use the three graph neural networks to learn node representations, and use them to train a node classifier for node classification.  ... 
arXiv:2107.01475v1 fatcat:qavrgjvxzbfzfg56x5ngxv3ioe

Variational Bayes in Private Settings (VIPS) (Extended Abstract)

James R. Foulds, Mijung Park, Kamalika Chaudhuri, Max Welling
2020 Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence  
Our framework respects differential privacy, the gold-standard privacy criterion.  ...  In the full paper we extend our method to a broad class of models, including Bayesian logistic regression and sigmoid belief networks.  ...  To solve this issue, graph neural network (GNN) has become a important way to help embed node structure information into graph node representation, especially for graphs of moderate size.  ... 
doi:10.24963/ijcai.2020/694 dblp:conf/ijcai/YanYH20 fatcat:pc4nelo7gzfmvmsiym3ohwspxa

NetFense: Adversarial Defenses against Privacy Attacks on Neural Networks for Graph Data

I-Chung Hsieh, Cheng-Te Li
2021 IEEE Transactions on Knowledge and Data Engineering  
Recent advances in protecting node privacy on graph data and attacking graph neural networks (GNNs) gain much attention. The eye does not bring these two essential tasks together yet.  ...  Imagine an adversary can utilize the powerful GNNs to infer users' private labels in a social network.  ...  Graph Neural Networks For the GNN models of f C and f P for semi-supervised node classification, we adopt Graph Convolutional Networks (GCN) [13] .  ... 
doi:10.1109/tkde.2021.3087515 fatcat:paojqkcf2rgvzagtxl5wiezi3y
« Previous Showing results 1 — 15 out of 4,996 results