Thesis/Project Final Exam Schedule

 

Final Examination Schedule

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

Autumn 2020
 

Tuesday, November 24

Aasveen Kaur

Chair: Dr. Min Chen
Candidate: Master of Science in Computer Science & Software Engineering

3:30 P.M.; Online
Project: ClickIndia: Indian Language Learning Mobile Application

The loss of native languages among the children of immigrants is a well-known roblem [1]. Yet, very little efforts have been made to protect them [2]. This project aims to build an iOS mobile application to facilitate the learning of two Indian languages, namely Punjabi and Hindi for English speaking children. The goal is to take advantage of smartphone technologies and constructivist principles of learning to allow app users to easily master commonly used Indian vocabulary words. This app will use location detection and multimedia presentation of the language content to create a location-based multi-modal learning environment. Further, it will use the novel concept of machine learning object recognition technology to develop an engaging language learning game. The proposed mobile application will have the following features:

  1. Rich representation of the language learning material using audio, text and pictures.
  2. Location-based local notifications to teach location-specific vocabulary, for example – User will be notified to practice animals/birds related vocabulary in Zoo premises.
  3. Category based vocabulary word list, for example – home, animals, etc. to build a connection between word and group for faster learning.
  4. Game-oriented learning to keep the user motivated.
  5. Use Machine learning model to identify user’s game responses, for example – In "Play to Learn" game, the app can automatically recognize the object user clicked in the picture.
  6. Crowdsourced picture tagging, i.e. the pictures clicked by app users will be stored on the cloud to build a tagged pictures dataset.
  7. A parent panel to provide feedback on the efficiency of the machine learning model.

Monday, November 30

Joseph Conquest

Chair: Dr. Michael Stiber
Candidate: Master of Science in Computer Science & Software Engineering

11:00 A.M.; Online
Project: Improving e-Science Workflow by Extending Provenance Visualization Functionality

Tracking data provenance is necessary for the domain of e-Science for the validation and replicability of experiments. Provenance is commonly coupled with scientific workflow management software to make it usable for scientists. Despite the prolific number of solutions and their diversity, a dilemma is still present if a scientist wishes to base their workflow on provenance, as the current solutions fail to provide a Graphical User Interface (GUI) based on provenance to guide workflow. BrainGrid Workbench ProVis provenance visualizer was developed, using system and data provenance visualization, to allow for the validation and creation of experiments. Using ProVis as the main control and view of the system highlights the importance of provenance in its ability to inform future experiments. Making provenance the focus for how to use scientific workflow systems enables the diverse requirements of e-Science workflows to be solved by this solution, as they have a common basis in provenance. Our study of usability shows a quick learning curve and easy deployment of experiments using provenance-based creation; as well as a reduction in the time necessary to create, build, execute, and analyze an experiment. 

Back to top

Thursday, December 3

Adriana Padilla

Chair: Dr. Kelvin Sung
Candidate: Master of Science in Computer Science & Software Engineering

11:00 A.M.; Online
Project: Location-based word recommendation system, modeled as a Multi-Armed Bandit problem

Personalized recommendation has garnered increasing attention in recent years. The most common dilemma that this problem faces is the balancing of "exploration” and “exploitation”, where “exploitation” selects content with known positive feedback, and “exploration” chooses new or less familiar content in order to better understand it. The exploration-exploitation trade-off can be effectively modelled as the Multi-Armed Bandit problem (MAB), which represents available options as bandits with corresponding rewards. Solutions for MAB problems aim to maximise the accumulated rewards through the balancing of exploitation and exploration of bandits.  

New language learning has been greatly facilitated by many mobile applications. Applications for vocabulary learning typically offer users a set of preselected words that can be learned through various engaging activities. An approach to further engage the user in word learning would be to leverage their physical location to recommend relevant words. Such a system can be modeled as a contextual Multi-Armed Bandit problem.  

This report details the modeling of a location-aware vocabulary recommendation application as a contextual MAB problem, and describes an approach to solving the problem based on the Thompson Sampling algorithm​.  ​In order to consider the location dimension, the Thompson Sampling algorithm is invoked to process words that are categorized according to the location of the device. Additionally, with the goal of providing users with a way to learn the recommended words, the application offers a set of vocabulary learning activities. The process of selecting words for the activities is also formulated as a recommendation system, where the user's familiarity with a word is the target of exploitation-exploration trade-off. 

Based on this system, over the course of three weeks, data was collected from a group of 12 testers who were asked to rate the relevance of the recommended words. Ratings were also compared between one user who started using the app in the first week of testing, and another who started after the three-week testing had ended. Overall, the results show that the system was able to weed out words with low relevance ratings, while repeatedly recommending those with high rating scores. As a result, it is observed that the average relevancy ratings had increased over time.  

These positive results demonstrate that the MAB algorithm can be effectively applied to solve the problem of location-based word recommendation, and that other context-based recommendation systems could also benefit from these algorithms. 

Friday, December 4

Agustinus Sutandi

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

11:00 A.M.; Online
Project: Detection of Traumatic Brain Injury Using Single Channel Electroencephalogram in Mice

Preclinical studies of traumatic brain injury (TBI) are often performed using a mouse model of mild traumatic brain injury (mTBI) due to highly controlled settings and high reproducibility in this experimental model, compared to studies of human TBI. Persistent changes in the sleep-wake cycle using a widely accepted mouse model of mTBI have been demonstrated. The gold standard of sleep-wake assessment is achieved by recording brain electroencephalogram (EEG), which not only allows for standard sleep staging but also allows further signal processing through quantitative EEG (qEEG) methods. The qEEG methods extract features from EEG recordings from different sleep stages that can further be used as an mTBI detection tool. The complex relationship among extracted features and different sleep stages in mTBI detection makes it a suitable problem to be solved using machine learning techniques. This capstone project implemented a supervised-learning system to detect mTBI and score sleep-or-wake stage from a single-channel EEG signal of duration less than 64 s by classifying the EEG signal into one of four classes: sham (control) wake, sham (control) sleep, mTBI wake, and mTBI sleep. Two approaches were implemented and compared. First approach used hand-crafted features extracted using well-known qEEG methods as inputs to a rule-based classifier such as k-nearest neighbor (kNN) and XGBoost. Second approach used deep convolutional neural network (CNN) to automatically learn features suitable for classifying the different classes. The highest cross-validated performance metrics achieved by the system was 80% accuracy and 79% average F1 (across target classes). The high performance of the system enables relatively quick and inexpensive detection of mTBI and sleep-or-wake state, which can have potential benefits for medical use.

Back to top

Tuesday, December 8

Eamon Maguire 

Chair: Dr. Marc Dupuis
Candidate: Master of Science in Computer Science & Software Engineering

5:45 P.M.; Online
Project: Key Invalidation in Peer-to-Peer Proof-of-Work Cryptocurrencies to Facilitate Creation of Key Replacement Framework and Prevent Fraudulent Transactions

As cryptocurrency transactions become more mainstream, we need to envision mechanisms to invalidate compromised private keys, and to grant new private keys to replace lost keys and associate them extant distributed ledger resources. Without a mechanism to prevent theft of distributed ledger resources, or a way to distribute new private keys for extant distributed ledger resources, it seems unlikely that distributed ledgers will gain widespread acceptance in government regulated commerce.

In an effort to take the first step towards a system to invalidate and replace compromised private keys, a mechanism to invalidate compromised private keys must be implemented and integrated with cryptocurrency systems. Once compromised private keys have been invalidated, additional extensions of the peer-to-peer blockchain implementation must be made in order to support issuing new keys. The issuance of new private keys necessitates a means by which we can associate an identity with blockchain resources, a means by which we can irrefutably prove that identity, and a means by which newly issued private keys can be bound to that identity's blockchain resources in a secure and irrefutable manner. These additional concerns, however, are beyond the scope of this work.

Wednesday, December 9

Yunbo (Robert) Chen

Chair: Dr. Min Chen
Candidate: Master of Science in Computer Science & Software Engineering

11:00 A.M.; Online
Project: Audio Chord Estimator – An Interactive Chord Estimation Web App using Machine Learning

Chord is a musical term that describes a tone made from several notes. This simple cumulation of notes renders the color of the song. The chord can be a simple strumming from an acoustic guitar, or the sum of notes played by bass, guitar, keyboard, synthesizer in a mixtape. Chord is hiding beneath the melody of the song and is essential for musicians. In a jazz concert, where improvisation happens very often, a player needs to know the chord progression of the music in order to play along and sound well.

Although chord exists in every song, it is hard to be identified, unlike melodies which one can easily sing out. Musicians must go through rigorous ear training to be able to identify the chord by ear. But for amateur music lovers, who just enjoy music but do not have a trained ear to know the chords, they need to go on the Internet and find piano sheet or guitar tabs.

Therefore, we have developed a web app that offers end-to-end chord estimation solution for music lovers to generate chord estimation of the song they like. By inputting the YouTube URL of the song or directly uploading the audio file, the application will use a self-trained chord estimation model to analyze the music and display an interactive result for every chord in the music. We put extra effort in the user experience so that user can anticipate the incoming chords on the page and play along with it.

Back to top

Questions: Please email cssgrad@uw.edu