Published:
Collection of notes from papers on the topic of training LLMs (and where possible, ML models in general) on edge devices. Most software libraries and literature aimed at on device operation are aimed at efficient inference. Training on device is a less-developed area. Training on device is useful in applications where a model must be fine tuned for individual use cases.
Published:
This article describes techniques to accelerate a Python codebase by exposing parallelized C++ functions using PyBind. It then analyzes the results of the optimization in which parallelizing one 40-line function in a 700-line program yielded up to a 3X end-to-end speedup.
Published:
Patent I authored while working at Microsoft. Describes an image segmentation technique for producing photorealistic human holograms.
Published:
Presentation slides for in-depth tech talk on dense 3D point cloud reconstruction from 2D images.
Published:
My most popular GitHub project: C++ and Python solutions to Cracking the Coding Interview.
Published:
Theory and implementation of Gaussian Mixture Models and the Expectation Maximization Algorithm.
Published:
Training a neural network to drive a car along a virtual track.
Published:
Training a neural network to classify traffic signs.
Published:
Creating a visual localization system for robot localization on Mars.