Faraz Rahman Math and Robotics @ University of Pennsylvania

About

Hello! I am an undergrad advised by Nikolai Matni in the GRASP Lab. During Spring 2025, I will be joining Dyna Robotics as an intern under Jason Ma.

I have been lucky to get my hands on

  • Generative Models
  • Optimization for Deep Learning
  • Controls for Robotics

Recently, I have been working on optimizers for representation learning with Thomas Zhang. Previously, I worked on early engineering at SwiftScore.

For fun, I play soccer, olympic lift, DJ, cook, and listen to all sorts of music.

Contact me at farazr [at] seas [dot] upenn [dot] edu.

Image

Projects

Optimizing for Representation Learning

Optimizing for Representation Learning

It's easy to forget that optimization and learning are not the equal. This project explores algorithms that can learn representations of data provably better than SGD.

Click to learn more
SwiftScore

SwiftScore

I worked on product design and engineering for SwiftScore - an AI EdTech Platform. My contributions include integrating Google Classroom, designing the Assignment Bank, and scaling the engineering team.

Click to learn more
Decision Making Diffusion Models

Decision Making Diffusion Models

Given a diffusion trajectory and some set of features in the final sample, can we ascribe a timeline for feature generation using some form of mechanistic interpretability? Considering the variance of a final sample of the diffusion process conditioned on intermediate points, we answer yes.

Click to learn more
TyPorch

TyPorch

This is a homebrewed C++ deep learning compiler that lets you design your own sequential models and run optimization algorithms such as stochastic gradient descent. Check out the repo for more!

Click to learn more

Blog

Quote Wall