wwstory

paper list

 
  • 2019年 10月01日

1.基础网络

ImageNet

net year detail download description
LetNet IEEE 1998 Gradient-Based Learning Applied to Document Recognition pdf CNN开山之作，手写体识别。
AlexNet ILSVRC 2012 ImageNet Classification with Deep Convolutional Neural Networks pdf ILSVRC 2012冠军，促进CNN发展。
VGGNet ICLR 2015 VERY DEEP CONVOLUTIONAL NETWORKS FOR LARGE-SCALE IMAGE RECOGNITION pdf 创建非常深的网络。
Inception CVPR 2015 Going Deeper with Convolutions pdf google设计，2014年的ImageNet冠军。
ResNet CVPR 2015 Deep Residual Learning for Image Recognition pdf 连接前后信息，可以训练更深的网络。
DenseNet CVPR 2017 Densely Connected Convolutional Networks pdf 与ResNet类似，所有层更加稠密的联系。
MobileNets 2017 MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications pdf  
DilatedConv ICLR 2016 MULTI-SCALE CONTEXT AGGREGATION BY DILATED CONVOLUTIONS pdf  

RNN

Unsupervised Learning

GAN

  • GAN 2014 Generative Adversarial Nets pdf 创新性思维，通过2个网络相互对抗的形式来训练，分别得到鉴别网络和生成网络。
    LSGANs 2016 Least Squares Generative Adversarial Networks pdf  
    WGAN 2017 Wasserstein GAN pdf  
    ** `` ** pdf  

Reinforcement Learning

2.优化

Model

  • Dropout 2014 Dropout: A Simple Way to Prevent Neural Networks from Overfitting pdf  
    Batch Normalization 2015 Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift pdf  
    Attention ICLR 2015 Neural Machine Translation by Jointly Learning to Align and Translate pdf 首次提出Attention，用于及其翻译，展示了attention对源语目标的对其效果，解释深度模型到底学到了什么。由于后续提出的概念，这个attention被称为soft/global attention。
    hard attention相关 ICML 2015 Show, Attend and Tell: Neural Image Caption Generation with Visual Attention pdf 提出hard/soft attention的概念，在图像上的应用。
    local attention EMNLP 2015 Effective Approaches to Attention-based Neural Machine Translation pdf 提出global/local attention的概念，对Attention的变化，其中multiplicative attention结构被广泛使用。
    self-attention `` Hierarchical Attention Networks for Document Classification pdf  
    transformer 2017 Attention Is All You Need pdf google提出的Transformer结构，完全摒弃递归结构，依赖注意力机制，可并行。（以往nlp中大量使用encoder-decoder结构，由于前后隐藏状态的依赖性，无法并行计算。）
    ** `` ** pdf  

Optimization

  • Adam ICLR 2014 ADAM: A METHOD FOR STOCHASTIC OPTIMIZATION pdf
    ** `` ** pdf

Loss Function

?

3.应用

NLP

Object Detection

  • - 2013 Deep Neural Networks for Object Detection pdf
    RCNN CVPR 2014 Rich feature hierarchies for accurate object detection and semantic segmentation pdf
    SPPNet 2014 Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition pdf
    Fast R-CNN IEEE 2015 Fast R-CNN pdf
    Faster R-CNN 2015 Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks pdf
    YOLO 2015 You Only Look Once: Unified, Real-Time Object Detection pdf
    SSD 2015 SSD: Single Shot MultiBox Detector pdf
    R-FCN 2016 R-FCN: Object Detection via Region-based Fully Convolutional Networks pdf
    Mask R-CNN 2017 Mask R-CNN pdf
    ** `` ** pdf

face

VQA

搜索