I'm a physicist working toward a physics of intelligence @ the Geometric Intelligence Lab at UC Santa Barbara. I'm especially interested in the geometry of representations inside neural networks, and the dynamics of learning.
In fall 2026, I am a researcher with the brain-computer interface (BCI) team at Science Corporation.
MIT '21 | PhD Candidate in Physics @ UC Santa Barbara | Advisor: Prof. Nina Miolane
I am a physics PhD candidate at UC Santa Barbara working at the intersection of deep learning and computational neuroscience. I study how neural networks build internal representations that support computation.
In my research I use dynamical systems theory, group theory, and geometry to understand emergent structure in deep learning systems. I am especially interested in how learned representations are shaped by the interplay between task structure, network architecture, and learning dynamics.
Prior to UCSB, I earned my S.B. in Physics from MIT in 2021. During my PhD, I have worked with Atmo as an AI researcher building deep learning models for large-scale atmospheric forecasting. I'm currently doing research on brain-computer interfaces at Science Corporation.
I also help organize NeurReps, the NeurIPS Workshop on Symmetry and Geometry in Neural Representations.
Feel free to reach me at facosta@ucsb.edu.
A Python package for Riemannian geometry in machine learning. Contributor since v2.6.0.
github.com/geomstats →Research code for sequential group composition experiments, related to the ICML 2026 paper.
github.com/geometric-intelligence/group-agf →