TensorFlow in Action: TensorBoard, Training a Model, and Deep Q-learning

by Sophia TurolMay 16, 2016
This blog post discusses TensorFlow’s mechanics of learning, overviews TensorBoard, as well as looks into Q-learning and convolutional networks.

videos-and-slides-from-tensorflow-meetup-in-london-march-2016

In data science, Q-learning represents an approach to learning about a state space and following the optimal policy thereafter. A classic example is a robot navigating a maze. Below are the videos from the TensorFlow London meetup featuring deep dive into TensorFlow’s architecture and a Q-learning example.

 

Technical overview of TensorFlow

In her session, Rebecca Murphy of Ocado Technology focused on TensorFlow under-the-hood mechanisms and demonstrated how to install the solution, load data, feed, checkpoint, and load the model. Then, she moved on to explaining what TensorBoard is and how to use it for visualizing learning. Finally, she talked about the further development of TensorFlow.

 

 

Below, you can check out the full slides by Rebecca.

 

Deep Q-learning with TensorFlow and PyGame

In his presentation, Daniel Slater from Bank of America Merrill Lynch touched upon such topics as reinforcement learning and Q-learning and how to apply these techniques to implement the PyGame framework with TensorFlow.

 

The PyGame solution described by Daniel Slater was also presented at the PyData conference earlier this year. In addition to a detailed description and great live demo, Daniel talked on TensorFlow, neural networks, reinforcement learning, Q-learning, and convolutional networks.

 

For more details, read a blog post on the matter by Daniel Slater or check out this project on GitHub.

 

Fireside chat

On delivering the presentations, Daniel Slater, Rebecca Murphy, and Peter Morgan shared their vision on:

  • What are the main challenges that can be solved with TensorFlow?
  • The recommendations for those getting started with TensorFlow
  • How can TensorFlow be further improved? How can the community contribute?

 

 

 

Join our group to get informed about the upcoming meetups!

 

Further reading

 

About the speakers

peter-morgan
Peter Morgan is a published author and computer science industry veteran with twenty years of experience in IT. Before entering the industry, he focused on high energy physics during his PhD studies at the University of Massachusetts and participated as Research Associate in Stanford University’s experiment that aimed to measure the mass of a neutrino. Peter is Cofounder and Technical Director of Data Science Partnership, where he currently oversees business development and helps clients to design and implement deep learning solutions.

 

Rebecca Murphy is Data Scientist at Ocado Technology, where she uses deep learning techniques to understand the intricacies of a customer choice. She has recently completed a PhD at the University of Cambridge, where she spent four years playing with high-power lasers and developed Monte Carlo methods to analyze data from fluorescently labeled protein molecules.

 

daniel-slater
Daniel Slater is an avid programmer who has worked across finance, computer games, and e-commerce. He is currently working as Assistant Vice President at Bank of America Merrill Lynch and doing his Master of Research in machine learning with a focus on reinforcement learning.
  •  
  •  
  •