자료유형  

 학위논문

Control Number  

0017161462

International Standard Book Number  

9798383191347

Dewey Decimal Classification Number  

004

Main Entry-Personal Name  

Zhang, Xiyuan.

Publication, Distribution, etc. (Imprint  

[S.l.] : University of California, San Diego., 2024

Publication, Distribution, etc. (Imprint  

Ann Arbor : ProQuest Dissertations & Theses, 2024

Physical Description  

210 p.

General Note  

Source: Dissertations Abstracts International, Volume: 86-01, Section: B.

General Note  

Advisor: Gupta, Rajesh K.;Shang, Jingbo.

Dissertation Note  

Thesis (Ph.D.)--University of California, San Diego, 2024.

Summary, Etc.  

요약The past decade has witnessed a significant growth of deployed sensors in our daily life, covering applications from healthcare, climate modeling to home automation and robotics. These sensors collect abundant time series data, which facilitate our understanding of various real-life processes. However, the inherent instability of these sensors, combined with the dynamic environments in which they operate, often results in the collection of data that is noisy, sparse, insufficient and varied. This presents a significant contrast to the data typically encountered in other domains such as computer vision and natural language processing. Consequently, current data-driven models trained under controlled experimental settings often fall short of their value in accurate inferencing and analyses. To address these limitations, we propose to bridge the inherent physical contextual knowledge and external semantic contextual knowledge of sensor time series to build a more robust analysis framework for sensor data. Specifically, we first exploit the inherent physics principles underlying time series data - ranging from mathematical models, spatio-temporal correlations to spectral properties - as inherent contextual knowledge. We leverage such physics knowledge to refine raw sensor time series and enhance data quality through denoising, imputation and augmentation. Additionally, sensor time series are often accompanied with external label names or metadata presented as text. Therefore, we build upon the recent advances in language modeling, to incorporate external semantic contextual knowledge from large language models or pre-train our own domain-specific foundation models. Such semantic knowledge further enriches sensor time series understanding and increases cross-domain robustness. Our framework is robust and demonstrates state-of-the-art performance in multiple tasks such as recognition, forecasting and navigation across sensing systems of various scales, from small-scale personal healthcare monitoring, smart home automation, to large-scale smart building control, energy management, climate modeling and beyond.

Subject Added Entry-Topical Term  

Computer science.

Subject Added Entry-Topical Term  

Information technology.

Subject Added Entry-Topical Term  

Robotics.

Index Term-Uncontrolled  

Sensors

Index Term-Uncontrolled  

Time series data

Index Term-Uncontrolled  

Energy management

Index Term-Uncontrolled  

Home automation

Index Term-Uncontrolled  

Climate modeling

Added Entry-Corporate Name  

University of California, San Diego Computer Science and Engineering

Host Item Entry  

Dissertations Abstracts International. 86-01B.

Electronic Location and Access  

로그인을 한후 보실 수 있는 자료입니다.

Control Number  

joongbu:655469