Study some papers

Instructions:

1. Spend at least 180 minutes (use a clock--no guessing) studying something that interests you about machine learning. Below are some papers that I think are worth studying. You may read these, or you may spend your time researching another topic in this field that interests you.
   
   
   • Adam: a Method for Stochastic Optimization. ADAM is a training method that is faster than stochastic gradient descent with deep neural networks.
   
   
   
   • Convolutional Deep Belief Networks for Scalable Unsupervised Learning of Hierarchical Representations. L¹ regularization makes deep conv nets form heirarchical representations.
   
   
   
   • Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. Batch Normalization is a special type of layer that makes deep neural networks train better.
   
   
   
   • Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks. GANs make two neural networks compete against each other such that the generative one produces realistic images.
   
   
   
   • Deep Residual Learning for Image Recognition. Skip connections can be used to form residual layers that learn faster by only learning the remaining difference with the target outputs.
   
   
   
   • On Intelligence. A guy describes a very interesting theory about how he thinks intelligence works. His theory focuses only on the cerebrum and ignores some other important aspects of intelligence, but overall I find it to be a very compelling theory. Here is a text version that you can use with AI Reader, or some other e-book reader while driving.
   
   
   
   • What are the computations of the cerebellum, the basal ganglia and the cerebral cortex?. Supervised learning, unsupervised learning, and reinforcement learning appear to be the primary learning mechanisms of three major regions of the brain.