A little past the anniversary of Neural Network Libraries 1.0’s release, we are now finally releasing Neural Network Libraries v1.1.0!

We have added double backward for most function layers, enabling computations of second-order derivatives!

Thank you for supporting Neural Network Libraries for the past one year, and please continue to support us!

Spotlight

Double Backward

Double backward, i.e., the second-order gradients of outputs with respect to inputs, is critical for implementing many state-of-the-art deep learning techniques.

We have enabled double backward for more than 70 function layers, highly enriching the applicability of Neural Network Libraries.

grads = nn.grad(outputs, inputs)

# Manipulate {grads} as usual variables.

Utilities

• Add setitem methods for NdArray and Variable
• Use module-wise random generator to initialize parameter weights

• Increase the map size for passing test

• Enable DataSource Shuffling During Training

• Improve ONNX and Tensorflow converter:

Bug Fix / Documentations

• [fix] device synchronization argument in mnist_runtime example

Examples

• Example: Wasserstein GAN-GP