Yi Li

Thanks for stopping by!

I am a fourth-year student at Boston College, double majoring in Computer Science and Management. Currently, I am on a gap year (Spring '23 - Spring '24) to deepen my engagement in academic research. I am applying for PhD programs starting in Fall 2024. I work as a research assistant at Ranger Lab under the guidance of Prof. Bryan Ranger, focusing on deep learning applications on ultrasound images. I have the privilege of collaborating with Prof. Maira Marques Samary on exploring the applications of LLMs in computer science education.

My research interests are centered around the development of trustworthy AI systems for healthcare and education, focusing on the ethical use of AI and building robust and reliable models to tackle existing problems.

Email: licds AT bc DOT edu

Publications

Leveraging LLMs and MLPs in designing a computer science placement test system

Yi Li, Riteng Zhang, Danni Qu, Maíra Marques Samary

[Paper] Accepted at CSCI 2023

Mining students’ mastery levels from CS placement tests via LLMs

Yi Li, Riteng Zhang, Danni Qu, Maíra Marques Samary

[Paper] Accepted at SIGCSE 2024 Student Research Competition

Ultrasound segmentation using deep learning: training on musculoskeletal phantom data and testing on clinical data

Yi Li, Keshi He, Hayoung Cho, Bryan Ranger

[Paper] Accepted at MIT URTC 2023

Projects

Comparative Analysis on Different Models with HC18 Challenge

Advised by Prof. Donglai Wei, we examined multiple variants of UNet such as Attention UNet and UNet++ on fetal head segmentation and attempted to replicate the best results in the HC18 competition.

[Paper] [Code]

MBTI Personality Classification with Performance Assessment

Advised by Prof. Emily Prud'hommeaux, we aimed to identify a person's MBTI personality type from the person's text. Our study found that logistic regression was most accurate in predicting the E/I and F/T personality pairs, with an accuracy rate of 58.4% and 62.1%, respectively. On the other hand, DistilBERT performed best in predicting the N/S and J/P personality pairs, with an accuracy rate of 73.8% and 71.6%, respectively.

[Poster] [Code]

Textual Emotion Detection with Deep Learning and Probablitic Approaches

Advised by Prof. Carl McTague, we implemented an interactive webpage to detect emotions (sadness, joy, love, anger, fear, surprise) from user input text, utilizing a dataset of 14,000 sentences for model training. Our best result is a 84.5% accuracy through a Bayes model with modified TF-IDF.

[Code]

Incremental Topological Sort

Advised by Prof. Hsin-Hao Su, I built and executed a testing program to examine the correctness and speed of the algorithm. The program expedites the previous testing approach by 12 times, providing valuable insights into the algorithm's performance and capabilities for further investigation.

[Code]

Selected Coursework

  • Computer Vision
  • Natural Language Processing
  • Computer Systems
  • Algorithms
  • Randomness and Computation
  • Logic and Computation
  • Swift iOS App Development
  • Computer Organization and Lab
  • Probablity
  • Intro to Abstract Math
  • Linear Algebra
  • Multivariable Calculus