Thesis/Project Final Exam Schedule

 

Final Examination Schedule

PLEASE JOIN US AS THE FOLLOWING CANDIDATES PRESENT THEIR CULMINATING WORK.

Summer 2020
 

Monday, June 22

Nasser Alghamdi

Chair: Dr. Munehiro Fukuda
Candidate: Master of Science in Computer Science & Software Engineering

11:00 A.M.; Online
Supporting Interactive Computing Features for MASS Library: Rollback and Monitoring System

Multi-Agent Spatial Simulation (MASS) library provides parallel execution over multiple computing-nodes for a wide range of agent-based simulations and data science applications. It conceals the complexity of parallel execution, dynamic and static entities allocation, and management process behind a set of APIs. Developing MASS applications for non-computing specialists or novices is a challenging and time-consuming task. Applications that depend on the library have to be compiled, distributed, executed for every change that is introduced by the  user. Additionally, the user has to use distributed log les or additional library calls for probing the application state. Thus, the user spends more time when experimenting on re-compiling, re-distributing, re-executing the application executable, and gathering information from distributed logs or results of additional calls. Though the library provides an intuitive programming model, its rigidity and the lack of convenient inspection tools can draw users away from using the library. In this project, we introduce InMASS (Interactive computing feature for MASS library) with two supporting features, namely: monitoring tool and rollback. We design computer experiment to emulate user-changes and we found that interactive version performs 9.2 times faster than non-interactive version when experimenting in ABM settings. Also, We compare and demonstrate how the  rollback and monitoring tool adds exibility and observability, respectively, to ABM systems by comparing InMASS against Repast Simphony, a well-known ABM framework.

Tuesday, July 28

Savitha Kumari

Chair: Dr. Brent Lagesse
Candidate: Master of Science in Computer Science & Software Engineering

3:30 P.M.; Online
Intuitive GUI based Context Instrumentation Framework

Crowdsensing is a technique where a large group of individuals having mobile devices capable of sensing and computing (such as smartphones, tablet computers, wearables) collectively share data and extract information to measure, map, analyze, estimate or infer (predict) any processes of common interest. AWARE is an open platform for context-aware mobile based crowdsensing research. This designed to capture, analyze and generate context aware data on mobile devices. The smart phones possess abundant sensors in recent times and can be used in multiple studies. AWARE is a solution to those researchers who are lacking the tools for mobile crowdsensing studies and are not capable to build one on their own. The toolkit abstracts the implementation details to the researcher and the participant. This allows them to focus more on sensor data collection and retrieval. This is enhanced with features like incentive mechanism for improving user participation, improved location privacy through obfuscation and data trustworthiness trackers based on user’s past contributions for better researcher satisfaction. Further the application is tested based on several functional tests and evaluated accordingly.

Wednesday, August 5

Jonas Pfab

Chair: Dr. Dong Si
Candidate: Master of Science in Computer Science & Software Engineering

11:00 A.M.; Online
Deciphering Protein Complex Structures from Cryo-electron Microscopy Maps using Deep Learning

Information about the macromolecular structure of protein complexes such as SARS-CoV-2, and the related cellular and molecular mechanisms can assist the search for vaccines and drug development processes. To obtain such structural information, we present DeepTracer, a fully automatic deep learning-based method for de novo multi-chain protein complex structure determination from high-resolution cryo-electron microscopy (cryo-EM) density maps. We applied DeepTracer on a previously published set of 476 raw experimental density maps and compared the results with a current state of the art method. The residue coverage increased by over 30% (45.65% to 76.93%) using DeepTracer and the RMSD value improved from 1.29Å to 1.18Å. Additionally, we applied DeepTracer on a set of 62 coronavirus-related density maps, among them 10 with no deposited structure available in the database. We observed an average residue match of 84% with the deposited structures and an average RMSD of 0.93Å. Furthermore, we found that the efficient structure modeling procedure of DeepTracer allows for exceptionally fast computations, making it possible to trace around 60,000 residues in 350 chains within only two hours. The web service is globally accessible at https://deeptracer.uw.edu.

Thursday, August 6

Meng Yang

Chair: Dr. Yang Peng
Candidate: Master of Science in Computer Science & Software Engineering

1:15 P.M.; Online
MoodPlay: An AI-Empowered Music Recommendation iOS App

Personalized and automatic music recommendation improves user experience, hence increases customer engagement and conversion. Representative music streaming service providers such as Spotify, Pandora, and Youtube Music all use various recommendation algorithms, often of a hybrid of Content-Based Recommendation and Collaborative Filtering, which study user’s listening behavior as well as the characteristics of the listened music. Though there exists work on the mood of music and human beings [18, 23, 38, 46, 48, 51, 56], application that recommends music based on user’s mood and activity is limited. The goal of this project is to create an iOS mobile application, called MoodPlay, which recommends music according to user’s mood and activity, estimated from real-time data collected by Apple Watch and iPhone. MoodPlay utilizes Machine Learning techniques to find the pattern between user’s mood and activity as well as user’s mood and a playing song’s tempo. Based on the discovered patterns, MoodPlay can recommend suitable songs to the user and further adjust its model in real-time according to user’s preference. By evaluating MoodPlay based on user reviews in App Store and the performance results of Machine Learning models and mobile application metrics, MoodPlay proves its high acceptance and usability.

Archit Gupta

Chair: Dr. Erika Parsons
Candidate: Master of Science in Computer Science & Software Engineering

5:45 P.M.; Online
Document Clustering Using a Fact Graph Approach

Words in human language are given meaning by their context and its structure. Sentences can vary in meaning depending on where punctuation is placed, while homonyms need additional information in order to be disambiguated. Bodies of text in human language are ordered and structured; sentences encode ideas, paragraphs group sentences, and chapters group paragraphs. However, it is often the case in the analysis of documents that we treat them as a bag-of words, i. e., a multiset of the contained words. In doing so, the meaning of the text derived from its structural properties is lost. We can attempt to preserve more of the documents' meaning by instead treating each of them as a graph of words, where the vertices are the words in the document, and the edges approximately represent relations between these words.  The relations are derived from the properties of the document's structure. In particular, we focus on using graph-encoded documents for clustering. The main motivation is to find an an alternative approach to the clustering algorithm LDA-GA. Data provenance reconstruction, which is concerned with approximating origins of documents within a dataset, uses LDA-GA for topic modeling, or discovering groupings of related documents [1, 2]. The original implementation of LDA-GA requires a large amount of system memory and resources, and results in significantly long execution times as well.  For these reasons, better performance is desired. In this capstone work, we apply graph-based clustering in order to group similar documents, and with the aim for better performance in terms of execution time and resource utilization. 

[1] Ailifan Aierken, Delmar B. Davis, Qi Zhang, Kriti Gupta, Alex Wong, and Hazeline U. Asuncion. A multi-level funneling approach to data provenance reconstruction. In e-Science Workshop of Works in Progress, October 2014

[2] Subha Vasudevan, William Pfeffer, Delmar Davis, and Hazeline Asuncion. Improving data provenance reconstruction via multi-level funneling approach. In the 12th International Conference on eScience, October 2016.

Thursday, August 13

Jessica Nguyen

Chair: Dr. Marc Dupuis
Candidate: Master of Science in Cybersecurity Engineering

11:00 A.M.; Online
Prioritizing Both Usability and Security: Closing the Feedback Loop Between UX Design, Software Development, Security Engineering, and IT/Operations

The agile development and information security industries have been pushing for the adoption of the DevSecOps model among enterprise technology companies in the past few years. The DevSecOps model is the result of multiple iterations of the software development life cycle (SDLC) as well as an increase in public interest, awareness, and research focused on the information security landscape. While this has been an important cultural and infrastructural shift for many technology companies, there is still a gap in this feedback loop that needs to be bridged: the gap between usability and security. Usability and security are often seen as competing qualities or trade-offs when developing a product. Normally, the development of security controls/features are seen as positive contributions to user and data protection. However, when these controls/features are not intuitive and difficult to use, both usability and security can be negatively impacted. This issue is widely known and acknowledged in the technology industry, yet the proposed solutions are largely vague, conceptual, untested, and unprioritized. Our work focuses on prioritizing both usability and security through further evolution of the DevSecOps model, introducing more cross-functional engagements between user experience (UX) designers, software engineers, and security engineers in a new proposed development model. This work covers the research methods for identifying the pain points in existing workflows for each team, the proposed solution workflows and tool suite, experimentation details, and the results from our experimentation. With the results of this experiment, we aim to quantify participants’ sentiments around a) usability of the proposed workflows and tool suite; b) interest in adoption; c) level-of-effort for adoption, and d) perceived value introduced by this model. The scope of this experiment is focused on the newly added engagements within the development model, and does not include end-to-end testing of all existing phases of the DevSecOps model. We acknowledge this as a limitation, and would be interested in providing this end-to-end experimentation coverage in future work.

Samantha Smith

Chair: Dr. Marc Dupuis
Candidate: Master of Science in Cybersecurity Engineering

5:45 P.M.; Online
A Study of Correlations Between Trait Affect and Phishing Susceptibility

Although phishing emails have been in use for decades, these social engineering attacks are still prevalent because they keep working; in fact, they are a leading cause of data breaches. In this research, I attempt to discern how an individual’s trait affect levels are related to their susceptibility to clicking on links in phishing emails, with particular attention on how this relationship may vary based on the type of phishing email employed. Trait Affect is a term from psychology that references a subset of one’s disposition and tendency towards certain moods and emotions. Trait Affect is further broken down into positive affect and negative affect, which are largely independent. Positive Affect reflects one’s tendency to act, while Negative Affect reflects one’s tendency towards experiencing negative emotions. Trait Affect has been shown to influence user’s behaviors and risk perception. Additionally, it is generally stable over an individual’s lifetime, making it a useful metric with which to model behavior. Being able to model an individual’s behavior in response to phishing is important to lowering the rates of phishing. While the creation of such models is outside the scope of this paper, the relationships examined will prove useful in future attempts to model such behaviors. To obtain data as close to a real-world scenario as possible, phishing susceptibility was measured on a click-through basis of emails sent to participant’s personal emails. This process caused some difficulty in managing to make emails that were both compelling and capable of passing automated email filtering. The process was further complicated by legal concerns surrounding the real-world approach to phishing. It is important to note, however, that no user data was taken – all measurements were based around a user clicking on a link rather than entering any information.  

Back to top

Questions: Please email cssgrad@uw.edu