A unified active learning framework for annotating graph data for regression task

被引:1
作者
Samoaa, Peter [1 ]
Aronsson, Linus [1 ]
Longa, Antonio [2 ]
Leitner, Philipp [3 ]
Chehreghani, Morteza Haghir [1 ]
机构
[1] Chalmers Univ Technol, Data Sci & AI, Gothenburg, Sweden
[2] Univ Trento, Trento, Italy
[3] Chalmers Univ Technol, Interact Design & Software Engn, Gothenburg, Sweden
基金
瑞典研究理事会;
关键词
Graph neural networks (GNNs); Active learning; Graphs-level regression; NETWORKS;
D O I
10.1016/j.engappai.2024.109383
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
In many domains, effectively applying machine learning models requires a large number of annotations and labelled data, which might not be available in advance. Acquiring annotations often requires significant time, effort, and computational resources, making it challenging. Active learning strategies are pivotal in addressing these challenges, particularly for diverse data types such as graphs. Although active learning has been extensively explored for node-level classification, its application to graph-level learning, especially for regression tasks, is not well-explored. We develop a unified active learning framework specializing in graph annotating and graph-level learning for regression tasks on both standard and expanded graphs, which are more detailed representations. We begin with graph collection and construction. Then, we construct various graph embeddings (unsupervised and supervised) into a latent space. Given such an embedding, the framework becomes task agnostic and active learning can be performed using any regression method and query strategy suited for regression. Within this framework, we investigate the impact of using different levels of information for active and passive learning, e.g., partially available labels and unlabelled test data. Despite our framework being domain agnostic, we validate it on a real-world application of software performance prediction, where the execution time of the source code is predicted. Thus, the graph is constructed as an intermediate source code representation. We support our methodology with a real-world dataset to underscore the applicability of our approach. Our real-world experiments reveal that satisfactory performance can be achieved by querying labels for only a small subset of all the data. A key finding is that Graph2Vec (an unsupervised embedding approach for graph data) performs the best, but only when all train and test features are used. However, Graph Neural Networks (GNNs) are the most flexible embedding techniques when used for different levels of information with and without label access. In addition, we find that the benefit of active learning increases for larger datasets (more graphs) and when the graphs are more complex, which is arguably when active learning is the most important.
引用
收藏
页数:25
相关论文
共 50 条
[41]   Active learning and semi-supervised learning for speech recognition: A unified framework using the global entropy reduction maximization criterion [J].
Yu, Dong ;
Varadarajan, Balakrishnan ;
Deng, Li ;
Acero, Alex .
COMPUTER SPEECH AND LANGUAGE, 2010, 24 (03) :433-444
[42]   Substep active deep learning framework for image classification [J].
Li, Guoqiang ;
Gong, Ning .
PATTERN ANALYSIS AND APPLICATIONS, 2021, 24 (01) :23-34
[43]   Active Learning Based Relation Classification for Knowledge Graph Construction from Conversation Data [J].
Ahmad, Zishan ;
Ekbal, Asif ;
Sengupta, Shubhashis ;
Mitra, Anutosh ;
Rammani, Roshni ;
Bhattacharyya, Pushpak .
NEURAL INFORMATION PROCESSING, ICONIP 2020, PT IV, 2020, 1332 :617-625
[44]   Graph-based Active Learning for Semi-supervised Classification of SAR Data [J].
Miller, Kevin ;
Mauro, Jack ;
Setiadi, Jason ;
Baca, Xoaquin ;
Shi, Zhan ;
Calder, Jeff ;
Bertozzi, Andrea L. .
ALGORITHMS FOR SYNTHETIC APERTURE RADAR IMAGERY XXIX, 2022, 12095
[45]   An active learning enhanced data programming (ActDP) framework for ECG time series [J].
Gupta, Priyanka ;
Gupta, Manik ;
Kumar, Vijay .
MACHINE LEARNING-SCIENCE AND TECHNOLOGY, 2024, 5 (03)
[46]   A Framework of Data Augmentation While Active Learning for Chinese Named Entity Recognition [J].
Li, Qingqing ;
Huang, Zhen ;
Dou, Yong ;
Zhang, Ziwen .
KNOWLEDGE SCIENCE, ENGINEERING AND MANAGEMENT, KSEM 2021, PT II, 2021, 12816 :88-100
[47]   Active Learning over Evolving Data Streams using Paired Ensemble Framework [J].
Xu, Wenhua ;
Zhao, Fengfei ;
Lu, Zhengcai .
2016 EIGHTH INTERNATIONAL CONFERENCE ON ADVANCED COMPUTATIONAL INTELLIGENCE (ICACI), 2016, :180-185
[48]   Semi-supervised Clustering Framework Based on Active Learning for Real Data [J].
Odate, Ryosuke ;
Shinjo, Hiroshi ;
Suzuki, Yasufumi ;
Motobayashi, Masahiro .
STRUCTURAL, SYNTACTIC, AND STATISTICAL PATTERN RECOGNITION, S+SSPR 2018, 2018, 11004 :184-193
[49]   Active learning framework for tunnel geological reconstruction based on TBM operational data [J].
Wang, Hao ;
Liu, Lixue ;
Shi, Maolin ;
Yang, Jie ;
Song, Xueguan ;
Zhang, Chao ;
Tao, Dacheng .
AUTOMATION IN CONSTRUCTION, 2024, 158
[50]   Graph Contrastive Learning with Constrained Graph Data Augmentation [J].
Xu, Shaowu ;
Wang, Luo ;
Jia, Xibin .
NEURAL PROCESSING LETTERS, 2023, 55 (08) :10705-10726