Deep Learning Papers

• 💡 Papers which gave me insights
• 📓 Categorized by problems and idea
• 📆 Sorted by chronological order
• 🔨 Clean implementations

Model Architecture

CNN

• Striving for Simplicity: The All Convolutional Net [arxiv, 14.12]
• U-Net: Convolutional Networks for Biomedical Image Segmentation [arxiv, 15.05]
• Deep Residual Learning for Image Recognition [arxiv, 15.12]
• Wide Residual Networks [arxiv, 16.05] [PyTorch]

Distribution Learning

RNN

• Generating Sequences With Recurrent Neural Networks [arxiv, 13.08]

GAN

Theories

• Generative Adversarial Networks [arxiv, 14.06]
• Energy-based Generative Adversarial Network [arxiv, 16.09]
• Wasserstein GAN [arxiv, 17.01] [TensorFlow]
• Boundary-Seeking Generative Adversarial Networks [arxiv, 17.02]
• BEGAN: Boundary Equilibrium Generative Adversarial Networks [arxiv, 17.03]
• Improved Training of Wasserstein GANs [arxiv, 17.04] [PyTorch]

Coarse-to-Fine Architectures

• Deep Generative Image Models using a Laplacian Pyramid of Adversarial Networks [arxiv, 15.06]
• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks [arxiv, 15.11] [TensorFlow]
• StackGAN: Text to Photo-realistic Image Synthesis with Stacked Generative Adversarial Networks [arxiv, 16.12]
• StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networks [arxiv, 17.10]
• Progressive Growing of GANs for Improved Quality, Stability, and Variation [arxiv, 17.10]
• High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs [arxiv, 17.11]

Learning Conditional Distributions

• Conditional Generative Adversarial Nets [arxiv, 14.11]
• InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets [arxiv, 16.06] [TensorFlow]
• Generative Adversarial Text to Image Synthesis [arxiv, 16.05]
• Adversarilly Learned Inference [arxiv, 16.06]
• Learning What and Where to Draw [arxiv, 16.10]
• Image-to-Image Translation with Conditional Adversarial Networks [arxiv, 16.11]
• Triple Generative Adversarial Nets [arxiv, 17.03]
• Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks [arxiv, 17.03]
• Learning to Discover Cross-Domain Relations with Generative Adversarial Networks [arxiv, 17.03]

Avoiding Mode Collapse to Learn Full Manifolds

• Unrolled Generative Adversarial Networks [arxiv, 16.11]
• Global versus Localized Generative Adversarial Nets [arxiv, 17.11]

VAE

• Auto-Encoding Variational Bayes [arxiv, 13.12] [PyTorch]

Representation Learning

• The Consciousness Prior [arxiv, 17.09]

Memory

• Neural Turing Machines [arxiv, 14.10] [PyTorch]
• Memory Networks [arxiv, 14.10]
• End-To-End Memory Networks [arxiv, 15.03] [PyTorch]
• Hybrid computing using a neural network with dynamic external memory [Nature, 16.10]

Continual Learning

• Overcoming catastrophic forgetting in neural networks [arxiv, 16.12] [PyTorch]
• Continual Learning with Deep Generative Replay [arxiv, 17.05] [PyTorch]

Attention

• Neural Machine Translation by Jointly Learning to Align and Translate [arxiv, 14.09]
• Show, Attend and Tell: Neural Image Caption Generation with Visual Attention [arxiv, 15.02]

Relational Reasoning

• A simple neural network module for relational reasoning [arxiv, 17.06]

Learning Equivariance

• Dynamic Routing Between Capsules [arxiv, 17.10]

Learning Invariance

• Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition [arxiv, 14.06]
• Spatial Transformer Networks [arxiv, 15.06]
• Deformable Convolutional Networks [arxiv, 17.03]

Learning Optimization

• Maxout Networks [arxiv, 13.02]

Model Optimization

• SplitNet: Learning to Semantically Split Deep Networks for Parameter Reduction and Model Parallelization [ICML, 17.07] [PyTorch]

Network Visualization

• Visualizing and Understanding Convolutional Networks [arxiv, 13.11]
• Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps [arxiv, 13.12]
• Understanding Deep Image Representations by Inverting Them [arxiv, 14.12]

Applications

Object Detection

• Rich feature hierarchies for accurate object detection and semantic segmentation [arxiv, 13.11]
• Fast R-CNN [arxiv, 15.04]
• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks [arxiv, 15.06]
• You Only Look Once: Unified, Real-Time Object Detection [arxiv, 15.06]
• SSD: Single Shot MultiBox Detector [arxiv, 15.12]
• YOLO9000: Better, Faster, Stronger [arxiv, 16.12]
• Fast YOLO: A Fast You Only Look Once System for Real-time Embedded Object Detection in Video [arxiv, 17.09]

Semantic Segmentation

• Learning Deconvolution Network for Semantic Segmentation [arxiv, 15.05]
• Fully Convolutional Networks for Semantic Segmentation [arxiv, 16.05]

Visual QA

• TGIF-QA: Toward Spatio-Temporal Reasoning in Visual Question Answering [arxiv, 17.04]
• A Read-Write Memory Network for Movie Story Understanding [arxiv, 17.09]