In this chapter, we continue our pragmatic exploration of the world of deep learning, analyzing two very important elements: deep convolutional networks and recurrent neural networks (RNN). The former represents the most accurate and best performing visual processing technique for almost any purpose. Results like the ones obtained in fields such as real-time image recognition, self-driving cars, and Deep Reinforcement Learning have been possible thanks to the expressivity of this kind of network. On the other hand, in order to fully manage the temporal dimension, it is necessary to introduce advanced recurrent layers, whose performance must be greater than any other regression method. Employing these two techniques together with all the elements already discussed in the previous chapter makes it possible to achieve extraordinary results in the field of video processing, decoding, segmentation, and generation.

In particular, in this chapter, we are going to discuss the following topics:

• Deep convolutional networks
• Convolutions, atrous convolutions, separable convolutions, and transpose convolutions
• Pooling and other support layers
• Recurrent neural networks
• LSTM and GRU cells
• Transfer learning