This MSS project focuses on developing AI models to isolate individual instruments or voices from live recordings of classical ensembles. Currently I am trying to use MIDI-generated and soundfont rendered datasets to pre-train and real-life recordings to fine tune the model. Ultimately, I aim to create tools that support human-AI collaboration in music—such as assisting with rehearsals, real-time accompaniment, and improvisation. Success in MSS will also unlock millions of classical recordings as training data, enabling AI to analyze and learn the intricate dynamics of musical collaboration.
The Music2Video project explores how AI can bring classical compositions to life through visual storytelling. Many classical works are rich with emotional and narrative depth, but they can feel inaccessible without a strong musical background. Drawing inspiration from my class project on music and moving pictures, I aim to develop AI systems that generate animations or atmospheric visuals tied to the structure and emotions of the music. This project is about making classical music more relatable and engaging, offering listeners a way to experience its beauty through another dimension of expression.
Energy-momentum properties, fundamental in classical and quantum physics, are challenging to define locally in general relativity due to the lack of Killing vector fields and the dynamical nature of spacetime under the equivalence principle. Physicists have proposed various approaches to address this issue. This thesis explores the problem's origins, reviews past solutions, proposes new solutions, and focuses on the Brown-York quasilocal energy-momentum tensor, detailing its derivation, applications, and limitations.
This work explores Symmetry Enriched Topological (SET) Order, focusing on symmetry fractionalization of anyon excitations. By calculating the strange correlator on a triangular lattice, I analyze temporal domain walls (gapless surface states without lattice symmetry breaking). Using the Grassmannian approach and Fourier diagonalization of the partition function, I find evidence of long-range correlations in these states.
One unique feature of Chinese, both Mandarin and its dialects, is the use of "tones"—continuous changes in pitch that distinguish meanings in spoken words. This tonal nature sets Chinese apart from most other major languages. How does this linguistic characteristic influence the composition and perception of melodies in Chinese pop songs? And what strategies might be developed to better integrate these tonal features into melody writing for a more harmonious blend of language and music?
