Opportunites for Research in Computer Science-Summer 2022

Dr. Shweta Ware

Studying ADHD using Smartphone Sensing and Ecological Momentary Assessment

The students have completed the following Citi training: Basic/Refresher Course - Human Subjects Research (Curriculum Group) UR Students conducting no more than minimal risk research (Course Learner Group) 1 - Basic Course
Prerequisite: CMSC 221 Data Structures.

Dr. Douglas Szajda

The Science of Adversarial Attacks on Automated Speech-to-Text

This research project is part of a continuing study to understand the ways in which automated speech-to-text systems “make decisions”. While the math of the deep learning machine learning models that underlie these systems is well understood, the systems themselves are large and complex, often with tens of millions of functional units (“neurons”) whose collective behavior can depend on billions of model parameters. It has been shown that even systems with impressive accuracy can be fooled –speech audio signals that are correctly transcribed can be modified such that a human listener perceives no difference in the modified audio, but the system outputs an incorrect transcription. Similarly, signals that sound to humans like generic background noise can be interpreted by the system as valid commands. Some of the most recent such attacks are poorly understood – attacks that seem to be only slight variations on successful attacks turn out to fail. In order to better understand how these systems “reason”, our study employs “explanation methods”, which are techniques designed to help researchers understand why deep machine learning models make the decisions that they do. The goal is to better understand the audio features that drive classifications (e.g., why is this audio segment interpreted as the “c” in “cat”), so that existing attacks can be better understood, and new attacks and defenses can
be identified.

Dr. Yucong Jiang

Data scraping and visualization for interdisciplinary research

Web scraping (e.g., of Congressmembers’ social media activity) to build data sets for analysis and visualization contributing to two interdisciplinary projects: respectively a computational social science project (conducting content analysis of political rhetoric) and a digital humanities project (building a digital archive of live music events). Students will learn and use web development languages (HTML/CSS/JavaScript) and selected popular libraries and tools.
Prerequisite: CMSC 221 Data Structures.

Dr. Jon Park

Machine learning/natural language processing

We’ve recently finished a project and are getting started on a new one on identifying argumentative structures and/or the process of persuasion.
The requirements are not hard and set, but in general:

- if you’re available this semester (0.5 unit independent study) and the summer: You should have completed Data Structures.
- if you’re available this summer only: You should have some experience with ML or NLP, as well.