Home Framework Chinese Researchers Introduce “FedPerGNN”: A Novel Federated Graph Neural Network (GNN) Framework for Efficient, Privacy-Preserving Personalization

Chinese Researchers Introduce “FedPerGNN”: A Novel Federated Graph Neural Network (GNN) Framework for Efficient, Privacy-Preserving Personalization

0
This Article is written as a summay by Marktechpost Staff based on the paper 'A federated graph neural network framework for privacy-preserving personalization'. All Credit For This Research Goes To The Researchers of This Project. Check out the paper and github.

Please Don't Forget To Join Our ML Subreddit

Machine learning is increasingly integrated into our daily lives without understanding. Our data is based on all the personalizations we have today, whether it’s ads, keyboard word suggestions, or other forms of individualized digital content. Personalization is an essential direction in web development. It can ease the stress of information overload by offering various services to different users based on their preferences and characteristics to better meet their needs. Current technology users are concerned about the privacy of their data and are reluctant to provide it to train these algorithms.

Storing raw data locally on user devices and developing local GNN models based on it is an attractive solution to address the privacy issue of these systems. However, in most cases, the data volume of consumer devices is insufficient to locally train accurate GNN models. A new machine learning algorithm, Federated Learning, has been developed to solve this problem. Federated learning is a privacy-preserving machine learning paradigm that can jointly develop intelligent models from data dispersed across many user clients while preserving privacy.

The Graph Neural Network (GNN) helps simulate high-order interactions and is commonly used in custom applications such as recommendations. However, due to the sensitive nature of user data, popular personalization solutions rely on centralized GNN learning on global graphs, which poses significant privacy concerns. Chinese researchers have developed FedPerGNN, a federated GNN system for efficient and privacy-preserving personalization.

Source: https://www.nature.com/articles/s41467-022-30714-9.pdf

Training of GNN models is performed cooperatively based on decentralized graphs inferred from local data using a privacy-preserving model update technique. To extend the use of graph information beyond local interactions, a privacy-preserving graph expansion technique is proposed that embeds higher-order information while preserving privacy.

FedPerGNN offers a viable approach to mining decentralized graphical data while protecting privacy for ethical and intelligent personalization. FedPerGNN generates 4.0% to 9.6% fewer errors than leading federated personalization algorithms under adequate privacy protection, according to experimental results on six datasets for personalization in various circumstances. Each client sends the locally computed gradients to a server for aggregation, and the aggregated gradients are then delivered to user clients for local updates.

The paper presents a privacy-preserving model update mechanism to ensure user privacy in model training, as transmitted model gradients may contain sensitive user information. Unlike previous solutions, which can protect private user ratings, this method can protect ratings and interaction histories, allowing for more extensive privacy preservation. Additionally, the solution does not require communication or local memory of the global element set, and its communication overhead is generally reasonable for today’s portable devices. FedPerGNN can be more easily applied in real-world personalization services because inferred local user item graphs only store lower-order interaction information. They present a privacy-preserving user-element graph extension protocol for extending local graphs and conveying higher-order information while preserving privacy. During this process, each client obtains anonymous user embeddings to expand the local subgraph, which facilitates the propagation of higher-order information on the user item graph while maintaining privacy to improve performance of the GNN model. The higher-order information on the user-item graph can be used successfully without incurring substantial communication costs after only a few privacy-preserving graph growth cycles.

Moreover, this method is not limited to the personalization scenario. It can be used as a fundamental strategy for privacy-preserving data mining on decentralized graph data, thus facilitating research in various fields involving graph-structured data. The code for this algorithm is available on GitHub.

With growing concerns about data privacy, algorithms like these help maintain data privacy while maintaining model accuracy.