BatchLayout: A Batch-Parallel Force-Directed Graph Layout Algorithm in Shared Memory

Abstract

Force-directed algorithms are widely used to generate aestheticallypleasing layouts of graphs or networks arisen in many scientific disciplines. To visualize large-scale graphs, several parallel algorithms have been discussed in the literature. However, existing parallel algorithms do not utilize memory hierarchy efficiently and often offer limited parallelism. This paper addresses these limitations with BatchLayout, an algorithm that groups vertices into minibatches and processes them in parallel. BatchLayout also employs cache blocking techniques to utilize memory hierarchy efficiently. More parallelism and improved memory accesses coupled with force approximating techniques, better initialization, and optimized learning rate make BatchLayout significantly faster than other state-of-the-art algorithms such as ForceAtlas2 and OpenOrd. The visualization quality of layouts from BatchLayout is comparable or better than similar visualization tools. All of our source code, links to datasets, results and log files are available at https://github.com/khaled-rahman/BatchLayout.

FusedMM: A Unified SDDMM-SpMM Kernel for Graph Embedding and Graph Neural Networks

“…For example, the original GCN implementation from Kipf and Welling [5] used SpMM to capture the graph convolution operation. SpMM-like operations were also used in high-performance graph layout [2] and embedding [4] algorithms.…”

“…Here, N (u) denotes inneighbors of u, ψ and φ are application-defined functions and is an application-defined aggregator. By changing the functions ψ, φ, and , we can easily derive forcedirected graph layout [1,2], graph embedding [3,4], graph convolutional network (GCN) [5], and graph neural networks (GNNs) algorithms [6] as shown in Fig. 1.…”

“…NOOP means no operation needed at that step 1. Needs a user-provided MLP function 2. A function that computes vector norms.…”

Force2Vec: Parallel force-directed graph embedding

“…In this paper, we recast the loss function in terms of a springelectrical model of force-directed graph drawing where two types of forces are calculated based on the connectivity of the vertices, namely, an attractive force when two vertices are connected by an edge and a repulsive force when there is no connection between a pair of vertices [20], [21]. With a good choice of the similarity function δ, the first term in Eq.…”

Training Sensitivity in Graph Isomorphism Network

“…Several unsupervised methods have been proposed in the literature to solve this problem sub-optimally [5,11]. Graph layout generation methods can also be used to generate embedding [12,13]. Some of them generate good quality embedding but have high runtime and memory costs while some of them consume less runtime and memory costs but generate moderate-quality embedding.…”