Joel Castro

I'm a fourth-year at UC Berkeley, majoring in computer science. My passions and interests lie in the realm of computer graphics, robotics, computer vision, simulations, and in particular, the overlap these arenas have with animation, film, video games, and AR/VR.

From: San Diego, CA
Age: 21
Pronouns: He/Him/His
Cat dad of: Milo, Mitchu, Sol, Rumi, and Cowboy

Download Resume Cat Showcaser

Publications

Research & Publications

Google Scholar | ACM Profile | ORCID

[Title TBD]

Authors: Austin Patel, Joel E. Castro Hernandez, Shuran Song
Venue: TBD
Status: Draft

[Abstract TBD].

Paper Coming Soon Conference Page Coming Soon

Associated Poster

View Poster Lightning Talk Presentation

Understanding Program Visualizations in the Wild

Authors: Joel E. Castro Hernandez, Olohi Goodness John
Venue: SPLASH, 2024
Status: Published

Visualizations play a significant role in writing, debugging, profiling, and generally understanding programs. However, little work has been done to understand the structure of program visualizations at a fundamental level— namely, why and how they scale. In this paper we present a theory of the interpretability of program visualizations focused on their abstraction and composition properties.

View Paper SPLASH Page ACM Digital Library

Associated Poster

View Poster

Selected for Poster & Oral Presentations at SPLASH
Selected for Poster Presentation at SACNAS

Interpretable biophysical neural networks of transcriptional activation domains separate roles of protein abundance and coactivator binding

Authors: Claire LeBlanc, Pooja Agarwal, Jack Demaray, Gean Hu, Marissa Zintel, Angelica Lam, Joel E. Castro Hernandez, Max Staller
Journal: Cell Press
Status: Under Review (Preprint)

Deep neural networks have improved biological prediction accuracy, but interpretation remains challenging. We designed simple neural networks incorporating biophysical models of transcriptional activation domains. These networks revealed how hydrophobic residues increase activation strength while decreasing protein abundance, and how acidic residues control both parameters. Combining biophysical and deep neural networks maximizes prediction accuracy and interpretability to yield insights into biological mechanisms.

View Paper bioRxiv NIH PubMed Page

Projects