Very Funny: Deep Learning of Humor from Gary Larson's Cartoons

Master Thesis

About This Topic

Build a deep learning application that is able to learn the humor of Gary Larson by analyzing the visual contents of the cartoons and the punchlines with deep neural networks. Tasks include pre-processing of an existing collection of 2000+ cartoons and categorization into N classes (hilarious, very funny, etc.), training of a convolutional neural network for the visual part, training of a recurrent/SLTM neural network for punchline analysis and of a third network/shallow classifier for merging. Requires prior knowledge and experience in the development of deep learning applications (e.g. Similarity Modeling lecture series or related courses from the Media Understanding module).