Reinforcement Learning with TensorFlow

Spatial features

Convolution Neural Networks are used to find hidden representations and is followed by applying attention mechanisms. Attention mechanisms direct the convolution layers of the network to focus on the relevant parts of the data. The advantage of using attention models is that it reduces the dimensionality of the dataset. As a result, a huge amount of computation, including convolution and so on, over the raw data is also reduced. 

The best approach to applying attention models is to use action and glimpse networks (explaining which is beyond the scope of this book but for further details in action and glimpse networks please go to this research publication "End-to-end Learning of Action Detection from Frame Glimpses in Videos ...