Generating Swipeable Tinder Profiles using AI: Adversarial & Recurrent Neural Networks in Realistic Content Generation
An implementation of StyleGAN and RNN architectures to generate profile biographical and image information, with the aim of generating believable female profiles as a proof of concept. All notebooks were ran in Google's Colaboratory environment with GPU acceleration enabled.
- StyleGAN pre-trained on FFHQ 1024x1024 dataset, cloned from NVIDIA's StyleGAN repository
- RNN trained on Kaggle Wine Reviews Dataset. Based upon the TrumpChar RNN project, with input from Brownlee's excellent RNN tutorials.
- RNN trained on the now-deleted Kaggle Tinder Female Profiles dataset. Legacy attempt kept for reference purposes, without access to dataset.
- RNN training notebook
- RNN text generation notebook
- StyleGAN image generation notebook
Full tutorial available here: https://medium.com/gradientcrescent/generating-swipeable-tinder-profiles-using-ai-adversarial-recurrent-neural-networks-in-dd68bd98c2f3