Deep Learning for Data-Driven Decision-Making

Zhiqiang Wan, University of Rhode Island


As we are entering the big data era, data are becoming high-volume, complex, and heterogeneous. Nowadays, almost every field and sector of the modern society is being impacted by big data, ranging from energy system to health care, and from business to government. The excessive amount of data contains potential and highly useful values. Extracting values from these data and utilize it to support the decision-making process is of paramount importance to improve productivity in business and enhance the well-being of our society. However, data-driven decision-making also arises with many challenges, such as feature selection, data fusion, and real-time decision-making. To overcome these challenges, this dissertation will develop deep learning algorithms and frameworks for data-driven decision-making. This work is composed of three major parts: automatic feature extraction, heterogeneous data fusion, and deep reinforcement learning (DRL) based data-driven frameworks. This dissertation will first investigate the automatic feature extraction to avoid the burden of the "feature engineering'' process. To make sure the feature extractor can be deployed to different scenarios, domain adaptive feature extraction will be studied. This dissertation will then examine the heterogeneous data fusion problem where feature refinement, normalization, transformation, and fusion will be investigated. With the automatic feature extraction and heterogeneous data fusion, this dissertation will develop data-driven decision-making frameworks for real-world applications, including robot-assisted pedestrian regulation and real-time electric vehicle (EV) charging management. Numerous experiments will be conducted to verify the effectiveness of the proposed frameworks.

Subject Area

Electrical engineering|Neurosciences|Engineering

Recommended Citation

Zhiqiang Wan, "Deep Learning for Data-Driven Decision-Making" (2020). Dissertations and Master's Theses (Campus Access). Paper AAI28149934.