/deep

Various deep learning algorithms implemented in TensorFlow

Primary LanguageJupyter Notebook

What's this?

Various deep learning algorithms implemented in TensorFlow.

Algorithms

MNIST

This is a tutorial for image classification using MNIST. 4 different implementation are provided in MNIST/mymodels, together with simple k-NN neighbor.

  1. mlp_notf.py
    • Multiple Layer Perceptron (MLP) without TensorFlow
  2. mlp.py
    • MLP with Tensorflow
  3. ???
    • Stacked Denoising Autoencoder + MLP
  4. cnn.py
    • Convolutional neural network (CNN)

The accuracy of those 4 (+1) programs can be evaluated on a same interface here.

Image Classification

Classification of CIFAR-10 by CNN and various preprocessing.

  • Data Augmentation
  • Global Contrast Normalization (GCN)
  • ZCA Whitening

Sentiment Analysis

Sentiment analysis of IMDb (movie review dataset) by Recurrent Neural Network (RNN).

Machine Translation

English-Japanese machine translation by Long Short-Term Memory (LSTM), trained by Tanaka Corpus.

Image Caption Generation

Image caption generation of MS COCO (Microsoft Common Objects in Context) by CNN + LSTM + Attention model.

Environment

Assume you are using anaconda3 environment.

pip install tensorflow-gpu
pip install keras