Can AI be used to generate funny memes?
This started out as a final class project for Machine Learning: Deep Learning at the Johns Hopkins University during the fall semester of 2020. We sought to improve upon previous work and ultimately settled on using a pre-trained GPT-2 language model from the Huggingface transformers library as the core driver of the meme caption generation. We fine-tuned on over 800 thousand meme captions we scraped from Imgflip. For more information on the machine learning underpinning the caption generation, check out our official GitHub repository.
This website was created by George Botev, and the machine learning pipeline powering the meme caption generation was developed by George Botev, Peter Ge, and Samantha Zarate.