Recommending Energy Retrofits at the Neighborhood Level

We use EnergyPlus™ simulations and neural networks to recommend energy efficiency retrofits across neighborhoods rather than just individual buildings.

Collaborators: Jorge Silveyra, Chetan Tiwari

Interpreting Regression Neural Networks with Linear Surrogates

I investigate when linear models fail to faithfully represent neural networks and propose the λ-score as a diagnostic; high surrogate fidelity ≠ accuracy.

Thesis: Detecting French Idioms

Alongside traditional linguistics, I apply back-translation and neural methods to a new French idiom corpus to improve our understanding of how computers and humans process idiomatic language.

Advisors: Sofia Serrano, Maria Hernandez

Toward a Grammar for French Idioms

This independent study in the French program is a major part of my thesis. It lays the groundwork by applying typical linguistic tools to process how idiomatic language functions in French. A major goal of the project is to produce something to the effect of a mathematical grammar which categorizes between idiomatic and non-idiomatic structures.

Advisors: Maria Hernandez