neon v1.1.0 released!

Highlights from this new release include:

* Sentiment analysis support (LSTM lookupTable based), new IMDB example network
* Support for merge and branch layer stacks via the introduction of LayerContainers
* Support for freezing layer stacks
* Adagrad based optimizer
* new GPU kernels for fast compounding batch norm, conv and pooling engine updates, new kernel build system and flags
* Modifications for Caffe support. Note that this may break backwards compatibility with previously serialized strided conv net models, see: http://neon.nervanasys.com/docs/latest/faq.html for details
* Default training cost display during progress bar is now calculated on a rolling window basis rather than from the beginning of each epoch
* Separate layer configuration and initialization steps
* Callback enhancements and updates. Note that validation_frequency renamed to evaluation_frequency
* Miscellaneous bug fixes and documentation updates throughout.

As always, you can grab this release from github at: https://github.com/NervanaSystems/neon

 

Scott Leishman
Algorithms Engineer