NAS uses Reinforcement Learning to search for convolutional network architectures.And the adversary to NAS, Aging Evolution for Image Classifier Search [4]..This strategy uses a modification of the genetic algorithm to search for convolutional network architectures.Both of these algorithms are able to design the architecture of CNNs that outperform humans..Compressing Neural NetworksTraining deep learning networks is a great way to get familiar with computer memory..Having the networks loaded into RAM memory enables much faster computing time.Research on compressing these networks such as Deep Compression [5] works very similar to JPEG image compression; Quantization and Huffman encoding..This has led to the development of architectures such as DCGAN [6], StackGAN [7], and Progressively-Growing GANs [8].ConclusionTo summarize, I think that Architecture Search, Compressing neural networks, and using GANs to build deep learning datasets are all going to be a fundamental part of the future of deep learning..Very Deep Convolutional Networks for Large-Scale Image Recognition..Neural Architecture Search with Reinforcement Learning..Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding..Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks.. More details