CNN (Convolutional Neural Networks) Abstract
CNN通常用來進行影象處理，經歷一系列卷積層,非線性層,池化層和完全連線層,最終得到輸出，輸出通常是一個單獨分類或者一組分類的概率.
Fully Connected Layer & Convolution Layer & Pooling Layer & Rectified Linear Unit
卷積層與全連線層的主要區別是卷積層保留了基本的空間結構(Spatial Structure)
全連線層(FC Layer)
Fully Connected
卷積層(Conv)
Conv通過filter輸出一個啟用對映(activation map) Conv經歷相較全連線來說主要有以下幾個特點:
- 區域性連線
- 保留空間結構
- 引數共享 (filter)
池化層(POOL)
同時也被叫做downsampling層．最受歡迎的就是最大池化層(max-pooling,如圖).
1. 減小輸入卷空間維度，也可以控制過擬合
2. 對於每個啟用層進行單獨處理
線性整流函式(ReLU)
在每個卷積層之後，通常會立即應用一個非線性層（或啟用層）。其目的是給一個在卷積層中剛經過線性計算操作（只是陣列元素依次（element wise）相乘與求和）的系統引入非線性特徵。通常意義下，線性整流函式指代數學中的斜坡函式，即
f(x) = max(0, x)
CNN Typical architectures
相關文章
- Convolutional Neural Networks（CNN）CNN
- 卷積神經網路：Convolutional Neural Networks(CNN)卷積神經網路CNN
- 深度學習之卷積神經網路(Convolutional Neural Networks, CNN)(二)深度學習卷積神經網路CNN
- 機器學習實戰（十三）：Convolutional Neural Networks機器學習
- 回顧一些重要的CNN改進模型（你真的瞭解 Convolutional Neural Networks 麼）CNN模型
- 吳恩達 Convolutional Neural Networks第二週quizzes吳恩達UI
- 卷積神經網路（Convolutional Neural Network，CNN）卷積神經網路CNN
- “卷積神經網路（Convolutional Neural Network，CNN）”之問卷積神經網路CNN
- Automatic Brain Tumor Segmentation using Cascaded Anisotropic Convolutional Neural NetworksAISegmentation
- 【論文翻譯】MobileNets: Efficient Convolutional Neural Networks for Mobile Vision ApplicationsAPP
- 論文閱讀—第一篇《ImageNet Classification with Deep Convolutional Neural Networks》
- 論文解讀二代GCN《Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering》GCASTZedFilter
- 論文筆記：Diffusion-Convolutional Neural Networks （傳播-卷積神經網路）筆記卷積神經網路
- 神經網路（neural networks）神經網路
- 深度學習論文翻譯解析（十七）：MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications深度學習APP
- 論文閱讀——Deformable Convolutional NetworksORM
- COMP9444 Neural Networks and Deep Learning
- DEEP LEARNING WITH PYTORCH: A 60 MINUTE BLITZ | NEURAL NETWORKSPyTorch
- 論文解讀（GIN）《How Powerful are Graph Neural Networks》
- 論文解讀（DAGNN）《Towards Deeper Graph Neural Networks》GNN
- Delphi 論文閱讀 Delphi: A Cryptographic Inference Service for Neural Networks
- Recurrent Neural Networks(RNN) 迴圈神經網路初探RNN神經網路
- Outrageously Large Neural Networks The Sparsely-Gated Mixture-of-Experts Layer
- 論文閱讀筆記：Fully Convolutional Networks for Semantic Segmentation筆記Segmentation
- pytorch實現 | Deformable Convolutional Networks | CVPR | 2017PyTorchORM
- DMCP: Differentiable Markov Channel Pruning for Neural Networks 閱讀筆記筆記
- 論文解讀（LA-GNN）《Local Augmentation for Graph Neural Networks》GNN
- 迴圈神經網路(RNN，Recurrent Neural Networks)介紹神經網路RNN
- ICLR2021-1：MULTI-HOP ATTENTION GRAPH NEURAL NETWORKSICLR
- 論文閱讀：《Deep Compositional Question Answering with Neural Module Networks》
- Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer 筆記筆記
- [論文閱讀] VERY DEEP CONVOLUTIONAL NETWORKS FOR LARGE-SCALE IMAGE RECOGNITION
- 遷移學習（DANN）《Domain-Adversarial Training of Neural Networks》遷移學習AI
- 論文解讀(GraphSMOTE)《GraphSMOTE: Imbalanced Node Classification on Graphs with Graph Neural Networks》
- 影像處理論文詳解 | Deformable Convolutional Networks | CVPR | 2017ORM
- 深度學習論文翻譯解析（十五）：Densely Connected Convolutional Networks深度學習
- 論文解讀（Geom-GCN）《Geom-GCN: Geometric Graph Convolutional Networks》GC
- （翻譯）DeepInspect: A Black-box Trojan Detection and Mitigation Framework for Deep Neural NetworksMITFramework