Easy-deep-learning-with-Keras

If you are unfamiliar with data preprocessing, first review NumPy & Pandas sections of Python for data analysis materials.

Materials in this repository are for educational purposes. Source code is written in Python 3.6+ & Keras ver 2.0+ (Using TensorFlow backend - For advanced topics, basic understanding of TensorFlow mechanics is necessary)

1. Multilayer Perceptrons

1) Basics of MLP

• Regression tasks with MLP
• Classification tasks with MLP

2) Advanced MLP - 1

• Weight initialization schemes
• Activation functions (Nonlinearity)
• Batch normalization
• Optimizers
• Dropout
• Ensemble of models

3) Advanced MLP - 2

• Putting it altogether

2. Convolutional Neural Networks

1) Basic CNN

• Basics of CNN architecture

2) Advanced CNN - 1

• Getting deeper with CNNs

3) Advanced CNN - 2

• CNN for sentence classification (imdb)

4) Using pretrained models

• Importing models already trained on ImageNet dataset (keras.applications)

3. Recurrent Neural Networks

1) Basic RNN

• Understanding RNN architecture
• Vanilla RNN (SimpleRNN)
• Stacked vanilla RNN
• LSTM
• Stacked LSTM

2) Advanced RNN - 1

• Deep RNNs
• Bidirectional RNNs
• Deep bidirectional RNNs

3) Advanced RNN - 2

• CNN-RNN

4) Advanced RNN - 3

• CuDNN LSTM
• CuDNN GRU

4. Unsupervised Learning

1) Autoencoders

• Autoencoder basics
• Convolutional autoencoder
• Dimensionality reduction using autoencoder

5. ETC

0) Creating models

• Sequential API
• Model Functional API

1) Image processing

• Importing images

2) Keras callbacks

• ModelCheckpoint
• EarlyStopping
• ReduceLROnPlateau

3) Using GPUs

• Make your training process faster with CUDA & CuDNN

4) Model selection

• Cross validation
• Grid search
• Random search

5) Class weighted learning

• Learning under class imbalance situations

6) Model weights

• Getting model weights
• Loading & saving model weights

6. Examples

1) Digit Recognition with RNN

• Simple RNN model
• Stacked RNN model
• Bidirectional RNN model
• Simple LSTM model
• Stacked LSTM model
• Bidirectional LSTM model
• Simple GRU model
• Stacked GRU model
• Bidirectional GRU model

2) Fashion item classification with MLP

• Simple MLP
• Autoencoder + MLP (dimensionality reduction)

3) Question generation with seq2seq (using Quora dataset)

 - Generating similar questions with seq2seq model

4) CNN for sentence classification

• CNN-static implementation of Kim 2014 paper

5) Sentiment Analysis in Korean (using Naver Sentiment Movie Corpus)

• Sentiment analysis with Logistic Regression (using sklearn & TF)
• Sentiment analysis with RNN

7. Text Analytics

Section with emphasis on text data analytics

1) Text processing

2) Word embedding

3) CNNs for text data

• 1-D Convolution for text analysis
• CNN for setnence classification (Kim 2014)
• Dynamic CNN for sentence modeling (Kalchbrenner et al 2014)
• CNN for text categorization (Johnson and Zhang 2014)