I study computers at the University of Michigan. I am working on some pretty exciting things and have a mild infatuation with physics.
Born and raised in Beijing, China, I moved to Montréal when I was 13
and then Toronto three years later. Now I'm a college student at the University
of Michigan studying Computer Science and a minor in Physics (?).
I can
fluently speak Chinese, English, and French. Some people that inspire me are
Kobe Bryant, Christopher Nolan, Friedrich Nietzsche.
Researching future of sustainable fair housing
Big Data Analytics, Research, Sustainable Development
The UN's eleventh sustainable development goal -- Fair Housing is an imminent issue. What features of housing organization are sustainable? How could urban planning be improved to address imbalances? In this research, we use Big Data to find the implications of the proximity to amenities of houses to their sustainability indices. This feature has been previously studied extensively vis-à-vis house pricing, but very seldom to demographic variables like educational attainment and equality.
In this research, I engineered a novel algorithm to granularly evaluate the proximity to amenities of houses. I designed the backend algorithm that, using Geoapify, a mapping API, queries all the nearby amenities to a house, then generating a proximity score for each category of amenities with an exponentially decaying proximity model. With a proportionally stratified simple random sample grouped by cities, more than 1600 communities across the United States are studied.
In addition to a new algorithm to acquire granular proximity data, this research also led to the conclusion that proximity to amenities is positively correlated with general educational attainment, and has no significant correlation with income levels or diversity. These findings provide valuable insights into current city planning and encourage further research into understanding the effects of housing proximity as a variable.
Turn your 5-second humming into an AI concerto
Generative AI, Music, Creative Continuation, Computer Vision, Gesture Recognition
While most generative AI models help create from scratch, vivAIdi explores the power in creatively continuing a work in progress, in the form of a 5-second humming.
Some of us have always dreamed of being a conducter, composing classical music masterpieces and conducting them. With the help of generative AI, this task has never been easier. Imagine with a few seconds of melodic humming and some simple gestures, you can conduct a whole orchestra to play music -- your original music! How much of a help would that be for a genre of music slowing dying out of the spotlight, and whose learners who are intimidated by its "complex" structures.
VivAIdi is built on electron.js as a local application. In the process of turning humming into a violin masterpiece, the input sound is converted to note sequences through the AutoCorrelate algorithm, then expanded with the MagentaAI(Tensorflow) framework. The process is then repeated for more instruments with SoundFonts applied to each of them to present a feeling of an orchestra. On the conducting end, a Tensorflow Computer Vision Fingerpose model is used to evaluate the gestures and command the sounds.
VivAIdi was conceived as a prototype at the MetHacks Hackathon in Toronto, and a full-version demo is in the works! Meanwhile, you can check out the Devpost project page or the GitHub repo linked below.
Gathering insights powered by graph-based DB
Data Visualization, Graph-based DB, Data Collection, Next.js
Introducing the 2023 Charged Up Scouting App, a powerful tool designed to greatly facilitate scouting at robotics tournaments. Built on a powerful Next.js front end and storing data in a graph database in Neo4j, this web application empowers scouts to record match data with ease and transform them into actionable insights -- data many degrees more profound than tabular data, recording teams' relationships with eachother. With intuitive radar charts, informative tables, and comparative bubble charts, teams can visualize performance metrics and strategize effectively for upcoming matches.
During crucial pick meetings, the Scouting App takes scouting to the next level by combining standardized statistics of robots into a power index, allowing the team to efficiently rank potential partners. This feature enables the formation of a comprehensive picklist, helping the team to identify the optimal alliance partners for maximum synergy and success. With plenty of aggregate, inferential, and relational data, every decision the team makes at tournaments can be statistically optimal, boosting the team's performance.