Projects
Here are some articles I (co)authored showcasing some of the projects I worked on.
|
|
Minimal Deepseek R1 inference
Robert Dyro
Open-source Deepseek R1 inference using JAX, minimal yet performant.
|
|
Optimizing LLM inference speed in float16 in JAX with Pallas
Robert Dyro
An attempt to improve the inference speed of HuggingFace's implementation of Mistral-7B in JAX for RTX 3090 via Pallas kernel tuning.
|
|
Pallas Puzzles
Robert Dyro,
Ed Schmerling
This set is puzzles is meant to teach you how to use Pallas from first principles in an interactive fashion. Adapted from Sasha Rush's Triton-Puzzles.
|
|
Post-training Compression of Neural Network via SVD
Robert Dyro
Post-training reduction of quantized NN memory footprint via SVD.
|
|
torch2jax - Call torch functions and models from JAX
Robert Dyro
A package for efficiently wrapping existing PyTorch code as JAX functions in a JIT-compatible way. AutoDiff gradients can be automatically defined, fully integrating any PyTorch code with JAX differentiation.
|
|
Automatic Grader
Robert Dyro
Automatic short answer grading with pretrained language models.
|
|
Optimal Control for Airplane Control
Robert Dyro,
Rohan Sinha
Controlling an airplane using optimal control with tuned objective parameterization using ray.tune.
|
|
GraphGym Tutorial and Neural Architecture Search
Robert Dyro,
Ricky Grannis-Vu
An introduction to GraphGym and an extension of GraphGym to perform Neural Architecture Search (NAS).
|
|
JAX Advanced Techniques
Robert Dyro,
Spencer Richards
This tutorial is at attempt to produce an educational reference for JAX techniques we found particularly interesting in our work and research. As such, it is a collection of a few disjoint topics.
|
|
Optimizing Models for Fairness and Explainability
Robert Dyro,
Somrita Banerjee
Optimizing Models for Fairness and Explainability via Shapley value and local linear model penalization and differentiation.
|
|
Sensitivity analysis using PyTorch and JAX
Robert Dyro,
Ed Schmerling
Optimization sensitivity analysis with PyTorch and JAX.
|
|
CUDA QP Solver
Robert Dyro
A QP solver fitting into a single CUDA block.
|
|
Sparse Automatic Differentiation
Robert Dyro
Reverse-mode automatic differentiation with full sparse Jacobians and Hessians.
|
|
Model Predictive Control
Robert Dyro,
James Harrison
Model Predictive Control - general and consensus optimization in Julia & Python.
|
Research
I'm interested in optimization, robotics, optimization, and
machine learning. My research is about creating new
models and algorithms for robotic systems that exploit known
structure of the problem.
|
|
Realistic Extreme Behavior Generation for Improved AV Testing
Robert Dyro, Matthew Foutter, Ruolin Li, Luigi Di Lillo, Edward Schmerling, Xilin Zhou, Marco Pavone
International Conference on Robotics and Automation (ICRA), 2025
Extracting learned behavior distribution for realistic counterfactual generation via efficient and scalable Hessian sketching.
|
|
Learning Deep SDF Maps Online for Robot Navigation and Exploration
Gadiel Sznaier Camps, Robert Dyro, Marco Pavone, Mac Schwager
arXiv, 2022
Optimal planning & control in new environments can be done effectively by learning deep signed distance function (SDF) maps from LiDAR data.
|
|
Second-Order Sensitivity Analysis for Bilevel Optimization
Robert Dyro, Edward Schmerling, Nikos Arechiga, Marco Pavone
Artificial Intelligence and Statistics (AISTATS), 2022
Differentiating through optimization can be done twice to obtain bilevel program Hessians.
|
|
Particle MPC for Uncertain and Learning-Based Control
Robert Dyro, James Harrison, Apoorva Sharma, Marco Pavone
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2021
Sampling arbitrary dynamical uncertainty and then optimization over the possibilities <em>jointly</em> gives safe real-time optimal control.
|
|