Reinforcement Learning with TensorFlow

Using a convolution neural network instead of a single layer neural network

Our gaming environment is videos and convolution neural networks have shown state-of-the-art results when it comes to computer vision. Moreover, the level of object detection in game frames should be close to human level ability and convolution neural networks learn representation from images similar to the way the primal visual cortex of humans does. 

DeepMind used three convolution layers and two fully connected layers in their DQN network that achieves superhuman level performance in Atari games as shown in the following flowchart: