Malarkey!

A powerful tool for description classification using generative adversarial networks.

Related Works • Learning Goals • Engineering Goals • UX Goals • License

Related Works

Learning to Generate Reviews and Discovering Sentiment is a great paper by OpenAI, which builds upon the principles used in Representation Learning (Bengio, 2013) in an exciting new way. These two papers inspired me to do this project.

Learning Goals

• Generative Recurrent Neural Networks
• Adversarial Classification
• Chose between PyTorch and TensorFlow

Engineering Goals

• Build an unsupervised generative network to create realistic tweets given a prompt
• Train a sister network to classify tweets using the feature vectors of the generative network
• Use Steven and a third sibling network to create an adversarial network pair by using both the generated data and real tweets

UX Goals

• Build a robust and user friendly way to collect human data on immeasurable features (ie. truth vs. lies, malarkey meter, funniness)

License

MIT

Atif Mahmud  ·  GitHub @Atif-Mahmud  ·  Twitter @Atif_Mahmud101