深度學習算法與編程 (暫停更新)

深度學習算法與編程

前言

本書內容

• 內容提要
  – https://blog.csdn.net/oBrightLamp/article/details/85162129

資料推薦

• 自學深度學習之計算機視覺的入門資料推薦
  – https://blog.csdn.net/oBrightLamp/article/details/84076410

開源許可 LICENSE

所有的說明性文檔基於 Creative Commons 協議, 所有的代碼基於 MIT 協議.
All documents are licensed under the Creative Commons License, all codes are licensed under the MIT License.

軟件版本

Python = 3.6
scikit-learn = 0.20.0
TensorFlow = 1.12
PyTorch = 1.0

損失函數

MSELoss

• 均方差損失函數MSELoss詳解及反向傳播中的梯度求導
  – https://blog.csdn.net/oBrightLamp/article/details/85137756

cross-entropy

• 通過函數圖像介紹信息熵的概念
  – https://blog.csdn.net/oBrightLamp/article/details/85269091

• 案例詳解cross-entropy交叉熵損失函數及反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/83962147

• Python和PyTorch對比實現cross-entropy交叉熵損失函數及反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/84029058

softmax

• softmax函數詳解及誤差反向傳播的梯度求導
  – https://blog.csdn.net/oBrightLamp/article/details/83959185

• 純Python和PyTorch對比實現softmax及其反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/84034658

softmax + cross-entropy

• 多標籤softmax + cross-entropy交叉熵損失函數詳解及反向傳播中的梯度求導
  – https://blog.csdn.net/oBrightLamp/article/details/84069835

• Python和PyTorch對比實現多標籤softmax + cross-entropy交叉熵損失及反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/84073485

優化算法

正則化 / 參數規範懲罰

• L2正則化Regularization詳解及反向傳播的梯度求導
  – https://blog.csdn.net/oBrightLamp/article/details/85290929

SGD, Momentum, RMSprop, Adam

• 常用梯度下降算法SGD, Momentum, RMSprop, Adam詳解
  – https://blog.csdn.net/oBrightLamp/article/details/85218783

• 純Python和PyTorch對比實現SGD, Momentum, RMSprop, Adam梯度下降算法
  – https://blog.csdn.net/oBrightLamp/article/details/85218799

Batch Normalization

• Batch Normalization函數詳解及反向傳播中的梯度求導
  – https://blog.csdn.net/oBrightLamp/article/details/84332455

• Python和PyTorch對比實現批標準化Batch Normalization函數及反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/84557854

• Batch Normalization的測試或推理過程及樣本參數更新方法
  – https://blog.csdn.net/oBrightLamp/article/details/85391056

• Python和PyTorch對比實現批標準化 Batch Normalization 函數在測試或推理過程中的算法
  – https://blog.csdn.net/oBrightLamp/article/details/85391167

特徵工程

PCA

• 特徵工程PCA降維方法的最大方差理論詳解
  – https://blog.csdn.net/oBrightLamp/article/details/85255895

• 純Python和scikit-learn對比實現PCA特徵降維
  – https://blog.csdn.net/oBrightLamp/article/details/85255898

全連接神經網絡

Affine

• affine/linear(仿射/線性)變換函數詳解及全連接層反向傳播的梯度求導
  – https://blog.csdn.net/oBrightLamp/article/details/84333111

• Python和PyTorch對比實現affine/linear(仿射/線性)變換函數及全連接層的反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/84453996

卷積神經網絡

convolution

• 卷積convolution函數詳解及反向傳播中的梯度求導
  – https://blog.csdn.net/oBrightLamp/article/details/84561088

• Python和PyTorch對比實現卷積convolution函數及反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/84589545

• 卷積convolution函數的矩陣化計算方法及其梯度的反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/85870773

• Python 實現 TensorFlow 和 PyTorch 驗證卷積 convolution 函數矩陣化計算及反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/85870813

Transpose Convolution

• TensorFlow和PyTorch對比理解卷積和反向卷積或轉置卷積(Transpose Convolution)
  – https://blog.csdn.net/oBrightLamp/article/details/85708124

MaxPool

• 池化層MaxPool函數詳解及反向傳播的公式推導
  – https://blog.csdn.net/oBrightLamp/article/details/84635346

• Python和PyTorch對比實現池化層MaxPool函數及反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/84635308

ReLU

• ReLU函數詳解及反向傳播中的梯度求導
  – https://blog.csdn.net/oBrightLamp/article/details/84326978

• Python和PyTorch對比實現ReLU函數及反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/84326804

dropout

• dropout函數詳解及反向傳播中的梯度求導
  – https://blog.csdn.net/oBrightLamp/article/details/84105097

• Python和PyTorch對比實現dropout函數及反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/84326091

循環神經網絡

RNN

• 循環神經網絡RNNCell單元詳解及反向傳播的梯度求導
  – https://blog.csdn.net/oBrightLamp/article/details/85015325

• 純Python和PyTorch對比實現循環神經網絡RNNCell及反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/85015402

• 純Python和PyTorch對比實現循環神經網絡RNN及反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/85015387

LSTM

• 長短期記憶網絡LSTMCell單元詳解及反向傳播的梯度求導
  – https://blog.csdn.net/oBrightLamp/article/details/85068285

• 純Python和PyTorch對比實現循環神經網絡LSTM及反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/85069255

GRU

• 門控循環單元GRUCell詳解及反向傳播的梯度求導
  – https://blog.csdn.net/oBrightLamp/article/details/85109589

• 純Python和PyTorch對比實現門控循環單元GRU及反向傳播
  – https://blog.csdn.net/oBrightLamp/article/details/85109607

生成對抗網絡

GAN

• 生成對抗網絡 GAN 的數學原理
  – https://blog.csdn.net/oBrightLamp/article/details/86553074

實戰

Kaggle

• PyTorch Kaggle 快速上手(雜草幼苗圖片識別)
  – https://blog.csdn.net/oBrightLamp/article/details/84947499

附錄

• 常用Latex公式
  – https://blog.csdn.net/oBrightLamp/article/details/83964331