Machine learning algorithms of the past were designed to capture the learning capabilities of the brain but with a high level of abstraction of its learning mechanisms. These abstractions resulted in shallow learning models that relied on hand crafted feature extraction on a problem specific basis with limited practical applications. The deep learning models of today represent the next generation of machine learning algorithms that can be trained from raw data using multiple processing layers due to novel modifications to the learning architecture compared to shallow learners. This development combined with the availability of raw data to train these models and the availability of fast affordable computers have enabled a great surge in its utility for many applications including video and audio pattern recognition. By exploiting the fundamentally different computing architecture and mechanisms prevalent in the brain, we believe that the next generation of machine learning called neuromorphic computing will advance the state-of-the-art in this field. In particular, it has the potential to realize energy efficient learning machines that could support a wide range of applications including internet of things, sensor processing, cybersecurity, robotics, mobile devices, diagnostics and prognostics and exoscale computing systems.