Portrait of Kenneth Tran

Kenneth Tran

Principal Research Engineer

About

I’m currently a Principal Research Engineer in the Deep Learning Technology Center. I have a wide interest in Machine Learning spanning from optimization algorithms to distributed systems.

My current main research pursuit is deep reinforcement learning with focus on

  • off-policy learning and sample efficient methods, safe exploration, reverse reinforcement learning
  • and real-world optimal control applications, including drones control, data center energy optimization, indoor farming optimization, etc.

Previously, I’ve also worked on

  • Convolutional neural nets and computer vision (object detection, image segmentation, image classification, etc.)
  • Image captioning and multimodal modeling
  • Optimization algorithms (SDCA, etc.)
  • DL/ML platform

In addition, I’m also the chief mentor of Microsoft AI School’s advanced projects class.

Projects

Deep Reinforcement Learning for Operational Optimal Control

Established: August 1, 2017

This research project aims at developing a new class of Deep Reinforcement Learning (DRL) algorithms and a DRL platform in the direction of off-policy learning, sample efficient methods, safe exploration, and reverse reinforcement learning. This set of algorithms and platform will be demonstrated in real-world operational optimal control applications such as Indoor Agriculture Optimization Data Center Energy Consumption Optimization Drones Optimal Control Etc.

Vision and Language Intelligence

Established: June 28, 2017

This project aims at driving disruptive advances in vision and language intelligence. We believe future breakthroughs in multimodal intelligence will empower smart communications between humans and the world and enable next-generation scenarios such as a universal chatbot and intelligent augmented reality. To these ends, we are focusing on understanding, reasoning, and generation across language and vision, and creation of intelligent services, including vision-to-text captioning, text-to-vision generation, and question answering/dialog about images and videos.

Any2Vec

Established: August 1, 2016

Embed objects of any modality (e.g. images, queries, sentences, etc.) to the same semantic vector space.

Publications

2017

2016

2015

2014

2009

Projects

Downloads

Microsoft?s fork of Caffe

December 2015

    Click the icon to access this download

  • Github