Generation models for internet of things sensing applications

Wang, Tianshi

Permalink

https://hdl.handle.net/2142/127254 Copy

Description

Title

Generation models for internet of things sensing applications

Author(s)

Wang, Tianshi

Issue Date

2024-12-03

Director of Research (if dissertation) or Advisor (if thesis)

Abdelzaher, Tarek

Doctoral Committee Chair(s)

Abdelzaher, Tarek

Committee Member(s)

• Nahrstedt, Klara
• Zhao, Han
• Srivastava, Mani

Department of Study

Siebel School Comp & Data Sci

Discipline

Computer Science

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Internet Of Things
• Generation Models
• Deep Learning

Language

eng

Abstract

The widespread deployment of Internet of Things (IoT) sensors has transformed how we observe physical phenomena and integrate computation into everyday life. However, despite the vast amount of data generated by these sensors daily, there remains a critical need for high-quality, task-specific datasets. As deep learning models increasingly demand larger volumes of data, the artificial generation of data for IoT sensing applications has become an essential research area. This dissertation presents a comprehensive exploration of generation models for synthesizing IoT sensing signals. Through seven chapters, it investigates the design, innovation, and practical application of various generative models in addressing key challenges in IoT data generation. Chapter 1 sets the stage by discussing the key challenges motivating the need for generation approaches in IoT sensing and outlines the scope. Chapter 2 demonstrates the effectiveness of discriminative models in generating signals with robust input-output mappings, exemplified by a task that transforms motion sensor signals into human speech audio. Chapter 3 delves into the strength of VAEs in disentangling the factors that influence data generation, highlighting a data augmentation framework for IoT applications as a proof of concept. In Chapter 4, diffusion models are explored, showcasing their capability to generate high-quality data in a vehicle detection scenario. Chapter 5 investigates the condition space of conditional generative models and the potential to manipulate this space for controlled data synthesis. Chapter 6 extends this exploration by proposing methodologies for achieving fine-grained control over the IoT sensing data generation process. Chapter 7 concludes this dissertation. This work advances the understanding of generation models in IoT contexts, providing innovative approaches to tackle the challenges of data scarcity and quality while paving the way for more intelligent and adaptable sensing applications.

Graduation Semester

2024-12

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/127254

Copyright and License Information

Copyright 2024 Tianshi Wang

Owning Collections