Machine Learning /
Primary contributor on the TFv2 branch for Machine Learning with TensorFlow: 2nd Edition, published by Manning and written by my mentor and Chief Technology and Innovation Officer (CTIO) at JPL, Chris Mattmann. [The first time my name appeared in a published book!]
Custom CUDA implementations for GPU-accelerated Gibbs sampling. Used in GMM, LDA, NLP, topic modeling, and more. [In collaboration with Tyler Hackett @ UCLA, and Matin Ghavamizadeh @ MIT]
Custom CUDA kernels for running convolutional neural network operations — i.e. convolution and general matrix multiply.
A model for predicting inference time given a set of convolution or general matrix multiply configurations and GPU architecture.
Field Programmable Gate Arrays (FPGAs)
USB-to-PS2 mouse controller for FPGAs written in Verilog. Performs clock division, signal sampling, processing, error checking, and validation. Includes Xilinx Basys 3 target configuration.
A custom built, half-precision FPGA neural network trained to recognize the MNIST data set. Displays the MNIST hand-written digits through a custom VGA controller and shows the predicted number on an onboard 7-segment display.
Ingest and validate EXT2 filesystem images. Performs consistency checks on binary data structures with various levels of indirection.
EXT2 Filesystem Audits — Performs block consistency, i-node allocation consistency, and directory consistency audits on a given EXT2 filesystem summary.