2/8/18

The next section of my course involves creating a structure called a convolutional neural network. Before deciding to dive in to the unit, I look up a brief introduction to convolutional neural networks on YouTube to determine whether or not they would be useful to my project. I quickly determined that they would not be; their primary use is image recognition/computer vision. I figured it would be good to go online and look for articles that discussed the use of machine learning in poker. One interesting tidbit of information I found that I had already inferred was the fact that poker is not a "perfect information game"; in other words, the AI players do not know exactly what cards the other players are working with, so there is a lot of unpredictability. Games like chess or GO are considered perfect information games because each player knows exactly what the opposing player has to work with. Next, I watched my course's introduction to convolutional neural networks just out of interest and decided for sure that they aren't what I'm looking for. After, that I moved on to Recurrent Neural Networks. I also watched a short film (starring Silicon Valley's Thomas Middleditch) in which actors performed a screen play written by a recurrent neural network. It was hilarious; the dialogue was nonsensical and made very little sense in context. Kind of reminded me of a bad lip reading video. Afterwards I skimmed an article talking about a poker AI called Libratus that has beat some of the best professional poker players; Libratus was trained using a technique called reinforcement learning, in which the AI plays games against itself to learn. I will continue to narrow down what method of learning I want to use in my project next class.