Hello, my name is Sam Fatehmanesh Vegas and I want humanity to conquer the light cone. I am a Physics major at Caltech in the Thomson Lab. While my interests range from wormholes to quantum computing, my current focus is on understanding the physics of intelligence and scaling human cognition. Feel free to read my thoughts or to check out my projects.

Projects

Generative Brain Model for Zebrafish

For the first time we have access to immense amounts of calcium imaging data for an entire brain. Utilizing this data we can pretrain foundation models similarly to how one trains LLMs on large amounts of text. The resulting large brain models are special since they act as low fidelity emulations, almost if not true uploads, of real zebrafish brains.

Optical Neural Interface (ONI) enabling Random Access Neurons

Calcium imaging serves as the gold standard for large scale high resolution measurement of neural activity. However, optically writing to neurons has hit a bottleneck. Thus, I am currently working on a novel technique to achieve high bandwidth and fully optical random access (read and write) for neurons.

CuNeuroCaRT

A toolkit and pipeline for taking large amounts of whole brain calcium imaging and extracting neuron masks, traces, and spikes. Powered by CUDA.

Human Super Intelligence (HSI)

Publications

From SGD to Spectra: A Theory of Neural Network Weight Dynamics

Brian Richard Olsen, Sam Fatehmanesh, Frank Xiao, Adarsh Kumarappan, Anirudh Gajula

ICML 2025 - International Conference on Machine Learning

View Paper

PILLARS: Plume-Deployed Inflatable for Launch and Landing Abrasive Regolith Shielding

Lily Coffin, AJ Torres, Stephanie Wallen, Pia Calub, Kevin D. Gauld, Isabella Kwaterski, Sam Foxman, Hannah Ramsperger, Emily Xu, Cesar Arellano, Belle Chen, Maria Gonzalez, Kieran Hale, Noah Howell, Carmen Quinones, Marilyn Recarte, Rithvik Musuku, Peijuan Huang, Louise-Marie Choi-Schattle, Marcel Liu, John Santana, Abdullah Almomtan, Claire Ellison, Hannah Rose, Sam Fatehmanesh, James Scott, Jasmine Wang, Kalind Carpenter and Soon-Jo Chung

AIAA Conference 2025

View Paper