Hyperparameter Tuning on Text Classification using CNNs (Convolutional Neural Networks);The Integration of Social Factors in the Development and Implementation of Autonomous Vehicles

Author:
Lei, Sarah, School of Engineering and Applied Science, University of Virginia
Advisors:
Seabrook, Bryn, EN-Engineering and Society, University of Virginia
Fleming, Cody, EN-Eng Sys and Environment, University of Virginia
Cohoon, Jim, EN-Comp Science Dept, University of Virginia
Abstract:

This Capstone project focuses on the applications of automated machine learning in text classification models. Text classification, the activity of labeling natural language texts with relevant categories from a predefined set, is a foundational task in many NLP (natural language processing) applications. These applications include sentiment analysis, web searching, and information filtering. By using text classifiers, companies can structure business information such as email, legal documents, web pages, chat conversations, and social media messages in a fast and cost-effective way. This allows companies to save time when analyzing text data, help inform business decisions, and automate business processes [1]. In this paper, we discuss the implementation of a model similar to Kim Yoon’s Convolutional Neural Networks for Sentence Classification. We then discuss the performance of hyperparameter tuning on the model for training optimization. Following, we propose additional ways to improve the performance of the model.
The STS research paper implores the integration of social factors in the implementation and development of AVs. Social demands of safety and ease have affected the development and implementation of AVs through modelling the technology around human decision-making. The STS research paper uses the Social Construction of Technology (SCOT) framework to analyze the definition and translation of social practices concerning safety and ease into implementable algorithms in the AVs. However, due to the extensive roles and relationships in the sociotechnical environment of AVs, several limitations exist that impede the successful realization of societal demands. Through the lens of the Wicked Problem framework, it is possible to then assess the complex network of social, economic, and political factors limiting proposed solutions for ethical decision-making in AVs. The examination of these limited proposals reveals a technologically deterministic future for AVs that include social, political, economic, and environment implications.

Degree:
BS (Bachelor of Science)
Keywords:
CNN, NLP, hyperparameter tuning, Wicked Problem
Notes:

School of Engineering and Applied Science
Bachelor of Science in Computer Science
Technical Advisors: Cody Fleming, Jim Cohoon
STS Advisor: Bryn Seabrook
Technical Team Members: Sarah Lei

Language:
English
Issued Date:
2020/05/06