<br /><br />The advantage of deep learning is that you don't have to engineer features. It extracts the features for you, which saves engineers months of work, and another claimed advantage of deep networks is that it requires fewer data to train. The third alleged feature is that the middle layers of deep networks can be redirected from one application to another. There are some compelling examples of this (“Analyzing the Performance of Layered Neural Networks to Identify Objects” is one example). This is likely a win over shallow grids because shallow grids do not have intermediate representations to speak of. Convolutional neural networks (CNNs), which consist of multi process layers to view data representations with multi abstract levels; it has been the most effective machine learning model in last recent years.<br />