Machine Learning Papers

As a personal challenge I summarized machine learning papers into a simpler, more friendly-form. Each paper has a link to the full summary.

Papers

• StyleGAN - New generator architechture for GANs that vastly improves performance and allows for a better understanding of latent space.
• Colorful Image Colorization - Approach to colourize gray scale images by treating the problem as a classification task.
• DRAW: RNN - Architechture that generates images more naturally sequentially
• Deep Residual Learning for Image Recognition - DL image classification framework to ease training and benefit from depth.
• First Order Motion Model for Image Animation - Framework to animate source images with motion derived from a driving video. Operates by extracting keypoints and affine transformations.
• Learning to Simulate Dynamic Environments with GameGAN - Generative Network to simulate environments only by ingesting screenplay and keyboard actions.
• Generative Adversial Networks - A new type of generative network. Two competing or adversial networks work against each other to improve each other.
• Cycle GAN - GAN used to translate image from source domain X to target domain Y without paired examples. Uses cycle loss to encourage translation bidirectionally.
• A Neural Algoirthm of Artistic Style - Synthesising style representation of one image and the content representation of another to form new images.
• R-CNN - A more precise image detecting algorithm that uses CNNs to extract features propose segmentation locations. Uses pretrained SVM for classification.
• Residual Attention Network for Image classification - CNN that uses the attention mechanism to increase performance.