Kyle Torres personal website and portfolio

Research

My research experiences have primarily focused on exploring the applications of statistical modeling to address problems in a variety of different fields.


Gibbs Sampling for LDA and Applications to RAG (Undergraduate Thesis)

In my thesis, I describe a method for deriving the posterior distribution used in Latent Dirichlet Allocation (LDA) and create a hybrid model in which I combine LDA with a retrieval-augmented generation (RAG) model. I find that this hybrid model outperforms a baseline RAG model in several areas including accuracy and processing time.

Write-up, Presentation


AI for Justice

At UCLA’s Computational and Applied Mathematics REU, I worked with faculty, peers, and the Innocence Center on a project developing a model to predict wrongful convictions using topic modeling algorithms such as non-negative matrix factorization, with all analysis conducted in Python.

Write-up, Presentation


Effect of ECMO Duration on Post-Transplant Survival

Through the USC Biostatistics and Data Science Summer Training Program, I collaborated with Biostatistics faculty and peers to analyze whether patients’ time spent on ECMO (life-support device) before heart transplant impacted post-transplant outcomes, applying models such as Cox regression and multinomial logistic regression in R.

Presentation, Poster


Analyzing Missing Data in the Stanford Open Policing Project

Through Pomona College’s Summer Undergraduate Research Program, I analyzed 23 traffic-stop datasets from the Stanford Open Policing Project with faculty and peers to assess bias from missing racial data, using odds ratios and R-based visualizations to evaluate potential imputations and reveal the impact of missingness on results.

Write-up, Poster