Google《图学习与挖掘》综述.pdf

发布时间：2022-05-31 发布人：admin 分类：说明书资料大小：21.87M 资料格式：pdf 举报版权申诉

syp_net-13633618-4744300845377764124.pdf-第1页.png

第1页 / 共312页

syp_net-13633618-4744300845377764124.pdf-第2页.png

第2页 / 共312页

syp_net-13633618-4744300845377764124.pdf-第3页.png

第3页 / 共312页

syp_net-13633618-4744300845377764124.pdf-第4页.png

第4页 / 共312页

syp_net-13633618-4744300845377764124.pdf-第5页.png

第5页 / 共312页

syp_net-13633618-4744300845377764124.pdf-第6页.png

第6页 / 共312页

syp_net-13633618-4744300845377764124.pdf-第7页.png

第7页 / 共312页

syp_net-13633618-4744300845377764124.pdf-第8页.png

第8页 / 共312页

文本预览

Mining and Learning with Graphs at Scale https://gm-neurips-2020.github.io/

Welcome + Agenda Introduction to Graphs + Application Stories What are graphs? Why are they important? Graph Mining: Basic tools and algorithms How do we build, cluster, and use graphs at scale? Graph Neural Networks How can we use deep learning on graphs? How can we use graphs in deep learning? Systems, Algorithms and Scalability How do we deal with massive graphs? How can graphs help us organize Google-scale data?

An Introduction to Mining and Learning with Graphs Vahab Mirrokni Graph Mining | go/graph-mining | December 2019

What are graphs? Graphs are representations of relationships (edges) between entities (nodes). In the most general case, graphs have: varying numbers of edges… - - with different edge types going to different node types… - with a highly complex structure. Social Networks Traffic, maps (Google Maps) Image Pixels Disease Spread https://www.pnas.org/content/116/2/401

Types of Graphs Natural graphs are graphs in which the edge relationship comes from an external source. Think: payments, social networks, roadways, coclick/cowatch. By contrast, similarity graphs are graphs in which the edge relationship is based on some measure of similarity/distance between nodes. In these cases, we start with a blob of (meta-)data and attempt to give that blob structure via graph representation.

Why Graphs? Computation on abstract concepts Most data is fundamentally about relationships, and graphs can help us . Graphs can also help us abstract local information and use it to extract useful global information from data. Computation on different data types We constantly deal with visual, textual, and semantic information, and all of this data relates to each other. Graphs provide a natural way to handle multi-modal data. Social Network Analysis, Wikimedia Commons

Search Query: Apple Why Graphs? Global and Local View Apple Inc. Global view: Graph structure/topology can tell us a lot about our data such as uncovering clusters of data points, or providing distance measures for otherwise intangible concepts. Local view: Local edges to and from a node can tell us something useful about a node -- something that is difficult to express with a single element. The black center pixel is part of an eye, but that is only apparent when you can see nearby pixels.

Graphs at Scale: Algorithms, Learning, & Systems for Impact Because graph representations are so flexible, we often want to use them on Google-scale data. We are often dealing with billions of nodes and many more edges. To work with data at this scale, we have to combine algorithmic ideas with the right systems and ML models. This can be very hard, and the devil is in details. These tools power hundreds of projects at Google in Search, Ads, Youtube, Play, Cloud, Maps, Payments, and more. Same-meaning queries for Keyword matching systems Better Caching for saving 32% Flash I/O for Search Infra(VLDB’19). Collaborative Filtering for YouTube Recommendations Finding micro-markets in designing A/B experiments [KDD’19, NeurIPS’19]

分享到：

赞收藏

资料库

Google《图学习与挖掘》综述.pdf

相关推荐

人工智能

热门标签

最新资料