Date of Award

6-10-2024

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Information Science

First Advisor

Mariofanna Milanova

Abstract

The advancements in machine learning, deep learning and AI have yielded remarkable tools and innovations, but certain groups face barriers preventing their utilization of these technologies. This research identifies and categorizes these barriers, focusing on three distinct groups: those lacking computational power, seeking to deploy models across multiple devices, and struggling with optimization challenges in high-performance computing centers. The study highlights the disconnect between academia's proposed solutions and their practical integration within industries and research centers, emphasizing the lack of convenience and integration among existing tools. To bridge this gap, this research offers a multifaceted approach. Firstly, it introduces publicly available tools and proposes a systematic guideline to facilitate the integration of these solutions, ensuring ease of use and applicability in various scenarios. Secondly, it investigates into specific objectives: optimizing deep learning models for energy efficiency, evaluating model compression with explainable AI techniques, and exploring hardware-based acceleration for deep learning inference in heterogeneous compute environments. The findings from these objectives are significant. This research demonstrates how video processing and classification tasks can be efficiently executed on personal devices through optimized deep learning models, eliminating the dependency on external cloud systems. Additionally, it highlights the limitations of traditional evaluation metrics in capturing the complexity of decision-making processes in AI models, proposing the incorporation of Grad-CAM for a more comprehensive evaluation approach. Moreover, the study addresses industry challenges in maintaining and evaluating compressed models for various hardware, proposing Grad-CAM as an effective post-compression evaluation method. Lastly, it explores hardware-based acceleration for deep learning inference in edge clusters, catering to the evolving needs of compute environments. Overall, this research not only identifies critical limitations hindering the widespread adoption of AI solutions but also offers tangible solutions and guidelines to empower users across technical and non-technical domains. Its implications extend to academia, industries, policymakers, and research centers, simplifying decision-making processes and reducing expenses related to computation, technology stacks, and deployment.

Download

Share

COinS