Sometimes it's painful to use TensorFlow. I recommend using Torch.
Reinforcement learning, for example, Atari games and Go, is just the cherry on cake. Although in our slide, this cake is a blackforest, which means that the cherry is indispensable. And it even has cherries inside.
The future is unsupervised learning.
There's no real progress made in deep learning in the decade except that we now use ReLU instead of Sigmoid.
