TensorFlow in Practice: Interactive Prototyping and Multi-GPU Usage

by Alex KhizhniakApril 27, 2016
Yaroslav Bulatov of Google demonstrated how TensorFlow works, providing recommendations and code samples for computational graphs, sessions, etc.

“Tricks of the Trade”

On March 31, 2016, Altoros organized the TensorFlow London meetup. At the event, Yaroslav Bulatov of Google presented some scenarios for using TensorFlow at scale—such as interactive usage, multiple GPUs, multi-threading, etc. In addition to an overview of TensorFlow’s architecture, Yaroslav also covered:

  • brief comparison to Theano, PyTorch, and NumPy (check out a more detailed analysis from us)
  • CPU/GPU and parallel execution
  • training parameters, assigning devices, etc.

At the end of the presentation, Yaroslav provided practical tips on interactive prototyping and demonstrated code samples. The slides are available here.


Want details? Watch the video!



Further reading

Performance of Deep Learning Frameworks