My Research
Researched parameter-efficient fine-tuning and smooth cascade unlearning through reversed self-distillation to make LLMs forget private data, showing that standard cascading between unlearning methods adds privacy risks.
Developed and implemented a type system and a type checker for a simple programming language in OCaml to ensure the noninterference property.
Conducted a study on table tennis ball trajectories, utilizing physics-based modeling and player data to analyze bounce uncertainty zones and strike timing, contributing to player classification and strategy development.
My Projects
Ongoing project with UC Berkeley and Satlyt, focusing on prototyping a ground-based system simulating Small Language Models deployment on satellites.
This project is an Android application prototype to improve students learning in classroom. We add functionalities to the open-source project NeutriNote, to evluate a student's note-taking and provide feedback for both the student and the professor.
Development of an algorithm to recognize words based on regular expressions and automata. From the user's input regular expression, the algorithm processes a file to determine if there is a match for the specified regular expression.
Automatization of the determination of an optimal color palette for an image by reducing the number of colors while preserving the original colors' representativeness.