Isaac Li

Incoming CS PhD at USC

San Jose, California, United States

About

Interested in theoretical computer science.

Experience

  • University of Rochester (2 yrs 2 mos)
    • StemForAll Program Instructor
      Jul 2025 - Aug 2025 · 2 mos

      Investigated the relationship between the learnability of a signal with its “Fourier Ratio,” the ratio between the L^1 and L^2 norms of its Fourier transform Proved that signals with low Fourier Ratio allow approximation via low-degree trigonometric polynomials, have low algorithmic rate-distortion and VC dimension, and display additive structure in high valued sets Wrote Python simulations to numerically estimate constants in Bourgain- and Talagrand-type inequalities

    • Vertical Integration of Research Lab
      Jul 2023 - Aug 2025 · 2 yrs 2 mos

      Mathematics research with Professor Alex Iosevich

    • Teaching Assistant
      Jan 2024 - May 2024 · 5 mos

      Design & Analysis of Efficient Algorithms Held recitation sessions and graded homeworks and exams covering topics including greedy algorithms, linear programming, dynamic programming, graph algorithms, divide and conquer algorithms, and P vs NP

  • Machine Learning Intern at Esperanto Technologies, Inc
    May 2024 - Dec 2024 · 8 mos

    Investigated the theory behind homomorphic encryption techniques (primarily CKKS) to evaluate feasibility for implementation and optimization on Esperanto hardware Studied approximate k-nearest neighbor algorithms for use in retrieval augmented generation

  • Machine Learning Intern at Esperanto Technologies, Inc
    May 2023 - Aug 2023 · 4 mos

    Implemented switch transformers on Vicuna by modifying the Llama architecture to add a Mixture of Experts (MoE) layer Analyzed RoBERTa parameter and activation distributions to investigate opportunities for quantization Led troubleshooting efforts after an AWS EC2 crash and found creative solutions to storage issues

  • Research Assistant at the Health & Environmental Economics Lab at University of Rochester Medical Center
    May 2022 - Aug 2022 · 4 mos

    Utilized parallel processing to significantly improve runtime of OCR data processing script Completed dataset which previously took 3 months in 23 hours