A Federated Graph Neural Network Framework for Privacy-Preserving Personalization

Graph neural network (GNN) is effective in modeling high-order interactions and has been widely used in various personalized applications such as recommendation. However, mainstream personalization methods rely on centralized GNN learning on global graphs, which have considerable privacy risks due to the privacy-sensitive nature of user data. Here, we present a federated GNN framework named FedGNN for both effective and privacy-preserving personalization. Through a privacy-preserving model update method, we can collaboratively train GNN models based on decentralized graphs inferred from local data. To further exploit graph information beyond local interactions, we introduce a privacy-preserving graph expansion protocol to incorporate high-order information under privacy protection. Experimental results on six datasets for personalization in different scenarios show that FedGNN achieves 4.0%~9.6% lower errors than the state-of-the-art federated personalization methods under good privacy protection. FedGNN provides a novel direction to mining decentralized graph data in a privacy-preserving manner for responsible and intelligent personalization.

Algorithms for Finding Shortest Paths in Networks with Vertex Transfer Penalties

In this paper we review many of the well-known algorithms for solving the shortest path problem in edge-weighted graphs. We then focus on a variant of this problem in which additional penalties are incurred at the vertices. These penalties can be used to model things like waiting times at road junctions and delays due to transfers in public transport. The usual way of handling such penalties is through graph expansion. As an alternative, we propose two variants of Dijkstra’s algorithm that operate on the original, unexpanded graph. Analyses are then presented to gauge the relative advantages and disadvantages of these methods. Asymptotically, compared to using Dijkstra’s algorithm on expanded graphs, our first variant is faster for very sparse graphs but slower with dense graphs. In contrast, the second variant features identical worst-case run times.

RDF Keyword Search by Query Computation

Keyword searches based on the keywords-to-SPARQL translation is attracting more attention because of a growing number of excellent SPARQL search engines. Current approaches for keyword search based on the keywords-to-SPARQL translation suffer from returning incomplete answers or wrong answers due to a lack of underlying schema information. To overcome these difficulties, in this article, we propose a new keyword search paradigm by translating keyword queries into SPARQL queries for exploring RDF data. An inter-entity relationship summary with complete schema information is distilled from the RDF data graph for composing SPARQL queries. To avoid potentially wasteful summary graph expansion, we develop a new search prioritization scheme by combining the degree of a vertex with the distance from the original keyword element. Starting from the ordered priority list that is built in advance, we apply the forward path index to faster find the top-k subgraphs, which are relevant to the conjunction of the entering keywords. The experimental results show that our approach is efficient and scalable.

Skeleton graph expansion of critical exponents in "cultural revolution" years

Kenneth Wilson's Nobel Prize winning breakthrough in the renormalization group theory of phase transition and critical phenomena almost overlapped with the violent "cultural revolution" years (1966–1976) in China. An unexpected chance in 1972 brought the author of these lines close to the Wilson–Fisher ϵ-expansion of critical exponents and eventually led to a joint paper with Lu Yu published entirely in Chinese without any English title and abstract. Even the original acknowledgment was deleted because of mentioning foreign names like Kenneth Wilson and Kerson Huang. In this article I will tell the 40-year old story as a much belated tribute to Kenneth Wilson and to reproduce the essence of our work in English. At the end, I give an elementary derivation of the Callan–Symanzik equation without referring to field theory.