{"title":"Learning with Small Data: Subgraph Counting Queries","authors":"Kangfei Zhao, Zongyan He, Jeffrey Xu Yu, Yu Rong","doi":"10.1007/s41019-023-00223-w","DOIUrl":null,"url":null,"abstract":"Abstract Deep Learning (DL) has been widely used in many applications, and its success is achieved with large training data. A key issue is how to provide a DL solution when there is no large training data to learn initially. In this paper, we explore a meta-learning approach for a specific problem, subgraph isomorphism counting, which is a fundamental problem in graph analysis to count the number of a given pattern graph, p , in a data graph, g , that matches p . There are various data graphs and pattern graphs. A subgraph isomorphism counting query is specified by a pair, ( g , p ). This problem is NP-hard and needs large training data to learn by DL in nature. We design a Gaussian Process (GP) model which combines Graph Neural Network with Bayesian nonparametric, and we train the GP by a meta-learning algorithm on a small set of training data. By meta-learning, we can obtain a generalized meta-model to better encode the information of data and pattern graphs and capture the prior of small tasks. With the meta-model learned, we handle a collection of pairs ( g , p ), as a task, where some pairs may be associated with the ground-truth, and some pairs are the queries to answer. There are two cases. One is there are some with ground-truth (few-shot), and one is there is none with ground-truth (zero-shot). We provide our solutions for both. In particular, for zero-shot, we propose a new data-driven approach to predict the count values. Note that zero-shot learning for our regression tasks is difficult, and there is no hands-on solution in the literature. We conducted extensive experimental studies to confirm that our approach is robust to model degeneration on small training data, and our meta-model can fast adapt to new queries by few-shot and zero-shot learning.","PeriodicalId":52220,"journal":{"name":"Data Science and Engineering","volume":null,"pages":null},"PeriodicalIF":5.1000,"publicationDate":"2023-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Data Science and Engineering","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1007/s41019-023-00223-w","RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, INFORMATION SYSTEMS","Score":null,"Total":0}
引用次数: 0
Abstract
Abstract Deep Learning (DL) has been widely used in many applications, and its success is achieved with large training data. A key issue is how to provide a DL solution when there is no large training data to learn initially. In this paper, we explore a meta-learning approach for a specific problem, subgraph isomorphism counting, which is a fundamental problem in graph analysis to count the number of a given pattern graph, p , in a data graph, g , that matches p . There are various data graphs and pattern graphs. A subgraph isomorphism counting query is specified by a pair, ( g , p ). This problem is NP-hard and needs large training data to learn by DL in nature. We design a Gaussian Process (GP) model which combines Graph Neural Network with Bayesian nonparametric, and we train the GP by a meta-learning algorithm on a small set of training data. By meta-learning, we can obtain a generalized meta-model to better encode the information of data and pattern graphs and capture the prior of small tasks. With the meta-model learned, we handle a collection of pairs ( g , p ), as a task, where some pairs may be associated with the ground-truth, and some pairs are the queries to answer. There are two cases. One is there are some with ground-truth (few-shot), and one is there is none with ground-truth (zero-shot). We provide our solutions for both. In particular, for zero-shot, we propose a new data-driven approach to predict the count values. Note that zero-shot learning for our regression tasks is difficult, and there is no hands-on solution in the literature. We conducted extensive experimental studies to confirm that our approach is robust to model degeneration on small training data, and our meta-model can fast adapt to new queries by few-shot and zero-shot learning.
期刊介绍:
The journal of Data Science and Engineering (DSE) responds to the remarkable change in the focus of information technology development from CPU-intensive computation to data-intensive computation, where the effective application of data, especially big data, becomes vital. The emerging discipline data science and engineering, an interdisciplinary field integrating theories and methods from computer science, statistics, information science, and other fields, focuses on the foundations and engineering of efficient and effective techniques and systems for data collection and management, for data integration and correlation, for information and knowledge extraction from massive data sets, and for data use in different application domains. Focusing on the theoretical background and advanced engineering approaches, DSE aims to offer a prime forum for researchers, professionals, and industrial practitioners to share their knowledge in this rapidly growing area. It provides in-depth coverage of the latest advances in the closely related fields of data science and data engineering. More specifically, DSE covers four areas: (i) the data itself, i.e., the nature and quality of the data, especially big data; (ii) the principles of information extraction from data, especially big data; (iii) the theory behind data-intensive computing; and (iv) the techniques and systems used to analyze and manage big data. DSE welcomes papers that explore the above subjects. Specific topics include, but are not limited to: (a) the nature and quality of data, (b) the computational complexity of data-intensive computing,(c) new methods for the design and analysis of the algorithms for solving problems with big data input,(d) collection and integration of data collected from internet and sensing devises or sensor networks, (e) representation, modeling, and visualization of big data,(f) storage, transmission, and management of big data,(g) methods and algorithms of data intensive computing, such asmining big data,online analysis processing of big data,big data-based machine learning, big data based decision-making, statistical computation of big data, graph-theoretic computation of big data, linear algebraic computation of big data, and big data-based optimization. (h) hardware systems and software systems for data-intensive computing, (i) data security, privacy, and trust, and(j) novel applications of big data.