Data Science, artificial intelligence and machine learning are driving much of the discussion around enhanced data analytics. Deep learning utilises neural networks with many hidden layers and is exceptional at identifying patterns in unstructured data, including image, sound, video and text data, as well as time-series data. These deep neural networks, however, require a large amount of data in order to set their weights correctly and can be slow to learn. In this presentation I will discuss the challenges to training deep neural networks, and how these can be overcome by using better design and utilising parallelisation offered by GPUs and horizontal scaling. 

The preview Azure Batch AI service enables scalable deep neural network training by maximising the compute power of CPUs or GPUs across multiple machines to minimise the challenges faced when training data models. Azure Batch AI supports an array of deep learning frameworks including TensorFlow, Caffe, Keras, and Microsoft’s own CNTK. This presentation will also examine the capabilities of Azure Batch AI and highlight use cases which can benefit from this scalable architecture.
Slide Deck 18 MB