Deep Learning Toolkit on the Data Science Virtual Machine


Deep learning is behind many recent breakthroughs in machine learning applications, including language translation with Skype, which was recently named one of the 7 greatest software innovations of the year by Popular Science, and speech recognition, where Microsoft recently achieved human-level parity at conversational speech.

On the 1st of Dec we will be launching our Azure GPU services.  These VMs combine powerful hardware (NVIDIA Tesla K80 or M60 GPUs) with discrete device assignment to enable powerful new options for training deep neural networks.

The deep learning toolkit for the DSVM is a solution for the Windows DSVM that installs several GPU-accelerated tools for deep learning, CUDA, cuDNN, the GPU driver and several samples. With the same steps required to create a new VM, you can have a DSVM ready for deep learning on Azure GPUs.

GPU-accelerated deep learning tools available with the VM include mxnet and CNTK, with more on the way.

A few samples are also installed in the C:\dsvm\deep-learning folder, including:

  • Character recognition on the MNIST dataset.
  • Image classification on the CIFAR-10 dataset.
  • Neural artistic style, a way to extract the style of an image and apply it to a new image.

These solutions clearly describe the elements involved in building a deep learning solution and demonstrate the power of the GPUs – use them to explore what’s possible or as a starting point for your own projects.


Microsoft Machine Learning Team Blog -

Learning Data Science