THIS IS A TEST INSTANCE. ALL YOUR CHANGES WILL BE LOST!!!!
...
Higher order gradient calculation is required for many applications such as adaptive learning rate optimization [1], WWGAN-GAN GP network [2], network neural architecture search [3], etc. Implementing higher order gradient can help unlock these applications and improve the usability and popularity of Apache MXNet framework.
...
[1] Forward and Reverse Gradient-Based Hyperparameter Optimization
[2] Improved Training of Wasserstein GANGANs
[3] Neural Architecture Search with Reinforcement Learning
...