Discussing TensorFlow History, Challenges, and Learning Perspective

by Sophia TurolFebruary 18, 2016
This blog post overviews under-the-hood mechanism of TensorFlow, features end-to-end examples, and discusses the future of the project as seen by Google Brain.

tensorflow-meetup-in-silicon-valley-january-2016-v11

TensorFlow is an open-source machine learning library originally developed by Google. The solution’s flexible architecture allows for deploying computation to one or more CPUs or GPUs in a desktop, server, or mobile device with a single API.

Watch the videos from the TensorFlow Silicon Valley meetup—sponsored and organized by Altoros on January 19, 2016.

 

TensorFlow overview: The learning perspective

In this video, Eric Danziger, Senior Engineer at a computer vision startup in San Jose, shared his experience of learning TensorFlow. He started with the MNIST demo and worked up to replicating parts of the “Playing Atari with Deep Reinforcement Learning” paper by Volodymyr Mnih et al.

 

 

An end-to-end example of using TensorFlow

In his talk, Delip Rao of Joostware focused on under-the-hood mechanisms of TensorFlow with an actual code example. His goal was to demonstrate various TensorFlow concepts in the context of a working application.

 

 

Fireside chat with the Google Brain team

This session with Yaroslav Bulatov and Lukasz Kaiser of the Google Brain team overviews the formation of TensorFlow in brief, provides some examples of the tool applied within Google products, plans for the future, etc. Highlights:

  • The history behind the project
  • Examples of TensorFlow behind the Google products
  • Problems that can be solved with TensorFlow
  • The most exciting thing about TensorFlow
  • The feedback from the external stakeholders and actions taken
  • How TensorFlow becoming open-source changed the roadmap
  • The plans for the future in terms of multi-GPU cloud deployments
  • How the “kubernetization” of TensorFlow happens
  • The DevOps aspect of TensorFlow: issues to handle
  • The problems TensorFlow is really good at solving or will be able in the future
  • The biggest challenges TensorFlow has right now and the resolutions to come in the next 3–6 months

 

 

Fireside chat: OpenAI and the future of deep learning

Gregory Renard, Chief Visionary Officer at XBrain—the company designing assistance for automakers, talked about OpenAI and the future of deep learning. (OpenAI is a non-profit artificial intelligence research organization founded by recognized machine learning/AI research engineers and scientists.) He highlighted the following aspects:

  • The problems the automotive / insurance industries face and how XBrain helps to solve them.
  • How and why such an organization as OpenAI can change the situation, where most of the practitioners in the deep learning are employed by a handful of companies.
  • The big trends in deep learning and how they are changing the future.
  • The industries that can benefit from the work of such an organization as OpenAI.

 

 

See you at the next meetups!

 

Further reading

 

About the experts

Eric Danziger is Senior Engineer at Cogniac, a computer vision startup in San Jose. He leads development of computer vision algorithms to run on the company’s cloud-based system. Eric has a bachelor’s degree in systems engineering from the University of Virginia and a master’s degree in robotic systems development from Carnegie Mellon University.

 

Delip Rao is Founder of Joostware. He is working on natural language processing and machine learning research problems (semisupervised learning, graph-based ranking, sequence learning, distributed machine learning, etc.) and published several highly cited papers in these areas. Prior to founding his own company, Delip was working at Amazon, Twitter, and Google Research.

 

Yaroslav Bulatov is Engineer at Google Brain, where he implements and trains large-scale neural networks. Yaroslav designed the first system that outperformed humans at recognizing outdoor house numbers. He is a graduate of Oregon State University’s School of Electrical Engineering and Computer Science, where Yaroslav researched hand recognition using geometric hand classifiers. Yaroslav also cowrote a paper on training conditional random fields—used in part-of-speech tagging, text-to-speech mapping, protein and DNA sequence analysis, as well as information extraction from web pages—via gradient tree boosting.

 

Lukasz Kaiser is Senior Software Engineer at Google Brain, where he works on machine learning and natural language processing (NLP), as well as constructs NLP-based translation and summarization systems. Before Google, Lukasz was a researcher in Paris, working on logic and the automata theory. He used satisfiability solvers and other symbolic methods for building game playing systems and program synthesis.

 

Gregory Renard is Chief AI Officer at xBrain. For over 20 years, Gregory has been working on data pipeline, unsupervised machine learning and deep learning applied to natural language processing, spoken dialogue systems, conversational platform, emotional AI, social robots, and voice search. As a former math professor and data scientist, Gregory is passionate about a connected world where universal access to information and knowledge frees people to achieve their full potential.