Navigation
About Me
Hi! I'm Emi Jhang, a recent graduate from Northeastern University in Boston, MA! My concentration lies in developing Machine Learning and AI programs/models, however I am well-versed in full-stack software development and have completed extensive projects in both domains. What interests me the most is application of ML/DL/AI in the medical field (such as early cancer detection, risk evaluation, etc.) as well as image processing and computer vision.
Work Experience
Griffiss Institute (AFRL)
AI/ML Co-op
The overall goal of my co-op project was to develop a way to automate the BDA process to aid analysts who do classification
manually. However, with the limited data available for training and testing, we cannot develop a reliable and
accurate model. The idea is to use generative image models to supplement the data with synthetic images in which we
test for the best performing model.
Such models included Pix2Pix, StyleGAN3, and Stable Diffusion. We evaluated
images using metrics such as CNNs, FID, and Siamese Networks to determine similarity to real images and whether
addition of images can outperform the model based in only real images. After confirming that supplemental
generated images have the capability to improve classification performance by over 10%, identified potential avenues
of future research and improvement.
Slides available here
Sanrel LLC
Software Engineering Intern
Sanrel LLC is a materials science start-up company. Alongside a team of interns, I collaborated to design, develop,
and test software for internal and client use. We set up a server and software to host an inventorying system for
lab materials and shipments. Additionally we reviewed, documented, and updated code for clarity and ease of use
for future engineers regarding projects to be taken up by Sanrel.
White Plains Hospital Center for Cancer Care
Lung Cancer Research Volunteer
At White Plains Hospital Center For Cancer Care, I helped a team of researchers by performing outreach and
scheduling for research patients. I screened applicants for study eligibility and communicated study
results thoughtfully. I organized materials for researchers and prepared documents, materials, and charts they would need
to complete patient labs and scans.
Projects
-
NextSongRec
Language/Tools: Python, PyTorch, Tensorboard, pandas
Recommended songs to add to a playlist/queue given a user's listening history. Using the Spotify Million Playlist Dataset and the 8M+ Song Audio Feature Dataset, modeled playlists as user listening sessions to recommend songs based on both ID and content features. With SASRec architecture as a baseline, achieved a hit rate of .83 and NDCG of .66. Modified TSSR architecture for our dataset and needs to evaluate on sequences of IDs and features rather than just ID and reached a conclusion of needing more comprehensive data to outperform the baseline.Slides available here
-
The Nexus
Language/Tools: Typescript, React, MongoDB, HTML, CSS, Jest, Javascript
Extended an existing codebase with a team to propose, develop and implement features encouraging community. Implemented a following and toast notification system for the existing “Stack Overflow” -esque application, as well as community spaces with real-time update polls using sockets and moderation capabilites. Additionally implemented private games and leaderboard that update in real-time as well.Live deployment available here
Github Repo available here
-
SatelliteMaskNet
Language/Tools: Python, PyTorch, Tensorboard, pandas
Developed a singular generalized NN model that performs pixel-by-pixel classification to segment buildings from a satellite image for analysts to work with to help determine factors such as building density and efficient land usage. The model was built to generalize to cities with various building densities and architectures and extract non-city specific features. On an unseen city achieves an accuracy of over 80%.Slides available here
-
Text Simplification
Language/Tools: Python, TensorFlow, nltk, transformers, word2vec, scikit-learn
Developed an NLP text simplification tool for the elderly and non-native English speakers. Created our own model using Word2Vec and tested against T5 Model and found our model outperformed T5 by human measure and readability scores. Program accessible through a React-Flask web application.Github Repo available here
-
NUConnect
Language/Tools: Python, MySQL, Flask, Docker
Prototyped a web app of a community forum for Northeastern students searching for co-op to connect with each other to reduce isolation during the searching process. Allows students and employers alike to post to a forum, apply to jobs, and see the status for applications submitted. Constructed a relational database maintaining 22 tables with 100+ attributes, with over 1k instances of sample data.Github Repo available here
-
ASL Image Classification
Language/Tools: Python, PyTorch, scikit-learn, pandas
Developed an image classifier for the ASL alphabet and numbers using CNN and multinomial logistic regression. Conducted extensive hyperparameter fine-tuning and augmentation to improve model robustness up to 97% accuracy with Pytorch-defined CNN.Report available here
-
Mushroom Classification
Language/Tools: Python, PyTorch, scikit-learn, pandas
Developed a machine learning program to classify mushrooms as poisonous or edible given a set of features. Trained and tested neural network, logistic regression, and k-NN classification models up to 99% accuracy. Accurately predicts edibility given either all 20 features, or only a few easily recognizable features.Report available here
-
Image Processing
Language/Tools: Java, JUnit, Swing
Coded a program that allows users to load and save edited images of file formats PPM, JPEG, PNG, and BMP. Used built-in Java GUI (Swing) to allow for an interactable interface and a model, view, controller design -
Marble Solitaire
Language/Tools: Java, JUnit
Using the terminal, users can play a Marble Solitaire Game that can be played in three different modes. Utilized a model, view, and controller design for implementation