arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2112.07768cs.LGcs.AI
收藏

大规模高效动态图表示学习

Efficient Dynamic Graph Representation Learning at Scale

Georgia Institute of Technology(佐治亚理工学院);Facebook(脸书);MBZUAI(穆罕默德·本·扎耶德人工智能大学)

Xinshi Chen, Yan Zhu, Haowen Xu, Mengyang Liu, Liang Xiong, Muhan Zhang, Le Song

摘要:

节点之间具有有序事件序列的动态图在现实世界的工业应用中很普遍,例如电子商务和社交平台。然而,动态图的表示学习由于数据的时间和结构依赖性以及不规则性而带来了巨大的计算挑战,从而阻止了此类模型部署到现实世界的应用中。为了解决这个问题,我们提出了一个有效的算法,有效的动态图形学习(EDGE),它选择性地表示一定的时间依赖性通过训练损失,以提高计算的并行性。我们表明,EDGE可以扩展到具有数百万个节点和数亿个时间事件的动态图,并实现新的最先进的(SOTA)性能。

英文摘要:

Dynamic graphs with ordered sequences of events between nodes are prevalent in real-world industrial applications such as e-commerce and social platforms. However, representation learning for dynamic graphs has posed great computational challenges due to the time and structure dependency and irregular nature of the data, preventing such models from being deployed to real-world applications. To tackle this challenge, we propose an efficient algorithm, Efficient Dynamic Graph lEarning (EDGE), which selectively expresses certain temporal dependency via training loss to improve the parallelism in computations. We show that EDGE can scale to dynamic graphs with millions of nodes and hundreds of millions of temporal events and achieve new state-of-the-art (SOTA) performance.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑