M.S. Data Analytics student at the University of Central Florida, with a background in music (B.A. Electronic Music, Bard College; M.A. Music Business, NYU).
My research focuses on AI-generated music. Specifically, can we tell whether a piece was made by AI, whom it imitates, and whether it copies existing works?
Projects
- AI-music-detection-unseen: I trained detectors on FakeMusicCaps and tested them on generators they had never seen. The CNN reached 97.5% balanced accuracy on known generators but caught only about half of the clips from an unseen one, which shows that a detector does not easily generalize to new models. (PyTorch)
- beethoven-recombination-detection: I reproduced a Beethoven composer classifier and tested it on imitations spliced from real Beethoven bars. The classifier often judged these imitations as Beethoven, while a detector that looks at the joins between bars could still find them. Both findings held when I repeated the tests on score MIDI files with real barlines. (Maestro, ASAP, PyTorch)
- airlink-intelligence: A course project (UCF CAP 6318) on how airport closures affect the US airline network. Some closures look minor nationally but cut off whole regions. It comes with an interactive website. (network analysis, Python)
- HittingtheVibe: A web app that builds a shared party playlist from guests' Spotify playlists. It started as an NYU course project, and I later developed it on my own. (Streamlit)
Website: xiangzongyou.com