Best Project Achievement

Scientific Concept Discovery: Using Machine Learning to Advance Scientific Research
Our group focuses on the question of how to design a learning framework that promote the generalizability of machine learning models. In this project, you will focus on exploring how neural networks acquire information from the training examples and how they learn to solve various physical problems (e.g., emulation of simple quantum systems). The premise of this project is that by observing how a machine learning model learns to solve the specific task, we can learn about the underlying problem itself. As an example, by analyzing the weights of a trained neural network, you can discover non-trivial symmetries of the modeled physical system, determine the relative importance of features, or identify some non-trivial interplay between underlying physical mechanisms. Your task would be to learn various tools for interpreting deep neural networks. You will test them in practice and you will explore methods that promote model transparency and interpretability.
Automatically segmenting and describing the human corpus callosum from brain MRIs
The human corpus callosum is the largest pathway connecting the left and right hemispheres of the brain. The shape of the corpus callosum (CC) changes throughout the course of human development, and it can also be altered with respect to disease onset. We can explore the variation in CC shape along the middle of the brain, but we need to extract it reliably first. The lab currently has two methods for extracting the CC, one using only image processing techniques, and another using deep learning (UNet) but these methods do not always extract the CC accurately. The accuracy results often depend on the MRI scanner that was used, or the abnormalities present in the scan. Can we improve the performance of our deep learning model with additional training data? Can we change some processing steps to improve the model? Once we do have an accurate segmentation, then what shape metrics of the CC as a whole, or in parts, are most telling of the underlying biology, such as age and risk for disease?
Automatically segmenting and describing the human corpus callosum from brain MRIs
Looking at White Hat (?) Hacker Social Networks on Github (Spring - 2021)
Open-source intelligence (“OSINT”) is a rapidly growing area of cybersecurity. This project seeks to explore OSINT information available on GitHub. Specifically, we will build and analyze a dataset comprised of users on GitHub who show a specific interest in GitHub repos related to hacking artifacts. This dataset and social network analysis could help us determine what attributes lead to “black hat” — or malicious — cyber actors.
Tracking health and nutrition signals from social media data (Spring - 2020)

This project will explore the ability to track real-life health and nutrition signals from social media data, focusing on data from Instagram and Foursquare. We will investigate the quality of Instagram posts as a source of data for measurements of dietary patterns and nutrition quality, focusing on spatial, textual, and (new in this semester) image content of posts linked to food outlets in Los Angeles, as well as nutritional content analysis of menus available online. Multiple aims will be investigated in this project, including: scraping data from social media; NLP of tag, comments, and menu data; image analysis; predictive models and social network analysis; and more. Also new in this semester: “ground truth” data on dietary patterns of LA residents will be available, enabling validation of dietary measures and predictive models built from Instagram posts.

The project will build on the DataFest 2019 project, and will expand the scope to actually access up-to-date data from Instagram, in particular: data with images, the underlying social connections / social network, and of course more timely (which requires data scraping).

Tracking health and nutrition signals from social media data (Spring - 2020)
Modeling Uncertainty in Drought Data
Droughts can have a substantial impact on agricultural systems and human livelihood. A Python package to calculate various drought indices in being developed. In this project, you will expand on this package and develop methods to test the sensitivity of the models to various input datasets and parameters.