Christopher J. MacLellan
Carnegie Mellon University
This project branched out of the final project for my SoftBots class, which was taught by Dr. Jeffery Van Baalen. In this project I wrote a simple python interface that downloads female pictures from www.hotornot.com. It then uses a facial detection algorithm to crop off everything but the face. The cropped face is then presented to the user who rates it as either hot or not. The whole time a machine learning algorithm (I tried both neural net and support vector machine) attempts to predict user ratings. The results turned out surprisingly well, I was able to do approx. 20% better then guessing their result. I view this significant but needing improvement. If one were to preprocess the images to normalize lighting, and/or possibly generate specific descriptors of the faces (hair color, nose type, proportions of face, etc...) and learn on these I believe the results would be amazing.