Samuel Kwong

About

In Summer 2020, I was a Software Engineer Intern at Google building deep neural networks for performance on TPU accelerators. I implemented deep learning models, ResNeSt and RegNet, in TensorFlow to run novel experiments for AutoML teams measuring throughput on TPU for direct comparisons to EfficientNet-X. Results include attaining a 74.7% throughput increase with ResNeSt porting from NVIDIA Tesla V100 GPU (float32) to TPU v3 (float32) and a 120% increase porting to TPU v3 (bfloat16).

In Spring 2020, I was a Software Engineer Intern at Verkada working on their backend and computer vision pipeline for enterprise video camera security. I pushed to production various improvements for facial recognition and vehicle detection and deployed a feature to mitigate occluded faces matching to each other as false positives (e.g. faces with masks).

My research interests are in the areas of computer vision, machine learning, and deep learning, with a focus on visual recognition and understanding of human actions in videos. As a computer vision researcher at Stanford Vision and Learning Lab, I have worked on video action recognition under the supervision of Juan Carlos Niebles.

Between 2006-2020, I was a competitive saber fencer. I won 3 U.S. National Championships in 2013, 2015 and 2016, and during my undergraduate years I represented the Stanford Varsity Fencing NCAA Division I team. In 2019-2020, I led the team as a captain, before graduating and retiring after Winter 2020.

Contact

samjkwong [at] gmail [dot] com

• LinkedIn
• Github
• Facebook
• Instagram

Elements

i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;
    deck.shuffle();
    i++;
}

print 'It took ' + i + ' iterations to sort the deck.';