top of page
Round ID Photo.png

ABOUT ME

I’m a PhD candidate at Yale University. My research examines how architectural constraints shape representational capacity in deep neural networks, and how this affects things like controllability and safety. To do this work, I combine mathematical analysis of model architectures with causal interventions on toy and trained systems.

 

Outside of research, I spend my time playing folk music, growing tomatoes, and enjoying the outdoors with my family. 

Selected Projects

1 / Architectural Constraints on Representational Ontology in Neural Audio Models

My dissertation project. The guiding question is: Do semantically meaningful musical features (e.g., pitch, melody, harmony) emerge as controllable internal variables in modern neural audio models?

2 / Musicological Interpretability with Generative Transformers

Analysis of embedding convergence and feature clustering in autoregressive transformers, examining how grammatical abstractions emerge under training.

3 / Play A Song From the 'Jukebox'

Early interpretability experiments on deep generative audio models, probing features and internal representation geometry. 

Nicole Cosme-Clifford

  • alt.text.label.LinkedIn

©2023 by Nicole Cosme-Clifford. Proudly created with Wix.com

bottom of page