Han, Zhu2022-06-30August 2022020-08August 202Portions of this document appear in: X. Du, H. Van Nguyen, C. Jiang, Y. Li, F. R. Yu and Z. Han, "Virtual Relay Selection in LTE-V: A Deep Reinforcement Learning Approach to Heterogeneous Data," in IEEE Access, vol. 8, pp. 102477-102492, 2020, doi: 10.1109/ACCESS.2020.2997729; and in: Du, X., Jin, Y., Wu, X., Liu, Y., Wu, X. (Sean), Awan, O., Han, Z. (2020, February 1). Classifying Cutting Volume at Shale Shakers in Real-Time Via Video Streaming Using Deep-Learning Techniques. Society of Petroleum Engineers. doi:10.2118/194084-PAhttps://hdl.handle.net/10657/10252Machine learning is one of the most promising fields of study nowadays. It is applied to various types of industry including image classification, object detection, and time-series signals prediction, etc. Latent space is a concept that is hidden but significant to machine learning, which helps extract features of data from different dimensions. In this dissertation, we try to apply machine learning with latent space representation and manipulation in real industrial applications both with two studies, respectively. We first apply machine learning with latent space representation to two works. First, we study the vehicle-to-vehicle relay networks with latent space to represent the decision of resource allocation in the reinforcement learning context. We propose a deep reinforcement learning model to decide the vehicles to be the relay. With the proposed model, the optimal decision is made and the largest overall data allocation is achieved. Then, we conduct a quantitative analysis of the cutting volume in real-time. This analysis is traditionally accomplished by workers on the rig, which cannot guarantee real-time and consistent reports of the cutting volume. With the proposed method, we are able to monitor the cutting volume in a real-time manner while relieving human labor. We then apply machine learning with latent space manipulation to another two works. First, we monitor the distribution of buffelgrass, a type of invasive grass based on the remote sensing images taken by unnamed aerial vehicles. By applying deep learning along with the discrete latent space-assisted data augmentation, the buffelgrass patterns are accurately located. Second, we solve a seismic inversion problem which is a workflow for deriving the subsurface model from seismic measurements. We propose to utilize autoencoder deep networks with latent space-aligned domain adaptation to migrate the trained model to unexploited data. With the proposed method, we prototype an inversion model with generalization capability quickly in a similar scenario.application/pdfengThe author of this work is the copyright owner. UH Libraries and the Texas Digital Library have their permission to store and provide access to this work. UH Libraries has secured permission to reproduce any and all previously published materials contained in the work. Further transmission, reproduction, or presentation of this work is prohibited except with permission of the author(s).Machine LearningLatent Space2022-06-30Thesisborn digital