Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


Jiongqian_dissertation.pdf (10.38 MB)

ETD Abstract Container

Abstract Header

Human-in-the-loop Machine Learning: Algorithms and Applications

Liang, Jiongqian
http://rave.ohiolink.edu/etdc/view?acc_num=osu1523988406039076

Abstract Details

2018, Doctor of Philosophy, Ohio State University, Computer Science and Engineering.
Machine learning is the process of learning meaningful patterns and extracting useful knowledge from data using computational and statistical techniques. While the overall goal is to help humans better understand the data and learn how to perform specific tasks, most of the current methodologies fail to fulfill it because they do not closely involve humans in the process and cannot apprehend human demands. This limitation has drawn more and more attention of researchers, and there is an increasing amount of study on human-in-the-loop machine learning. However, we are still facing a series of challenges in this area, especially 1) to fully capture the supervision and demands of the humans; 2) to effectively incorporate human supervision into machine learning; 3) to scale up the human-guided machine learning to large-scale datasets. In this dissertation, we show how to tackle these challenges on a few particular machine learning tasks and present some promising results towards this direction. First, we discuss how to conduct outlier detection robustly and efficiently with contextual information provided by humans. In this problem, we ask users to select some of the attributes as context and study anomalous behavior with the rest of attributes. Second, we show methods to mine relationships in heterogeneous information networks (HINs) following the interests of the humans. In this problem, the users provide the pair of targeted entities and the type of relationship that they are interested in, and we propose a novel method (called PRO-HEAPS) to efficiently discover most suitable relationship instances from the information networks. Third, we investigate how humans can be involved in a practical medical data analysis framework for a game-based stroke rehabilitation system, called RehabANLYS. Specifically, we study how to integrate the domain knowledge from doctors with the participation of patients to provide effective analysis on rehabilitation measurements. Fourth, we perform flood mapping on satellite images by leveraging guidance from humans. We study how to incorporate the expertise from domain experts to conduct quality flood mapping (HUG-FM). In addition, we develop a crowdsourcing platform and try to use the wisdom of the crowd for flood mapping (CHUG-FM). Fifth, we offer a more advanced method for the flood mapping task by leveraging deep neural networks. For this purpose, we propose SEANO for performing embedding on attributed networks with outliers, which has generic applications for graph mining beyond flood mapping. Finally, we seek to speed up existing graph embedding techniques by using a multi-level paradigm (called MILE). We repeatedly coarsen the graph and conduct graph embedding on the coarsest graph, from which we refine the embeddings to the original graph through a graph convolution model. Through these research efforts, we emphasize the importance of human involvement in machine learning and study how to incorporate human guidance into the machine learning process while ensuring the scalability of the algorithms. The long-term goal of the research in this dissertation is to facilitate the notion of human-in-the-loop machine learning and stimulate research to realize this idea on more machine learning tasks.
Srinivasan Parthasarathy (Advisor)
Arnab Nandi (Committee Member)
Huan Sun (Committee Member)
Guo-Liang Wang (Committee Member)
251 p.
Computer Science
machine learning; human-in-the-loop; human guidance; outlier detection; graph mining; graph embedding

Recommended Citations

Citations

  • Liang, J. (2018). Human-in-the-loop Machine Learning: Algorithms and Applications [Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1523988406039076

    APA Style (7th edition)

  • Liang, Jiongqian. Human-in-the-loop Machine Learning: Algorithms and Applications. 2018. Ohio State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1523988406039076.

    MLA Style (8th edition)

  • Liang, Jiongqian. "Human-in-the-loop Machine Learning: Algorithms and Applications." Doctoral dissertation, Ohio State University, 2018. http://rave.ohiolink.edu/etdc/view?acc_num=osu1523988406039076

    Chicago Manual of Style (17th edition)

osu1523988406039076
2,271
© 2018, all rights reserved.
This open access ETD is published by The Ohio State University and OhioLINK.