Sequential Data GAN

---

This module is to generate sequential data with GAN implemented by LSTM.

The basic structure is followed ckmarkoh's git https://github.com/ckmarkoh/GAN-tensorflow

Used Data

For making sequential data, MNIST 0-3 data are used (0->1->2->3).

Modeling

Each modules, genrator and discreminator are designed with 2 layer LSTM and 1 layer Fully Connected Network. Generator is designed as one-to-many model, which get one random vector as input, and generates sequential images. Discriminator is designed as many-to-one model, which get sequential images, and decides that is real or fake ones.

Results

Epoch 10

Epoch 30

Epoch 50

Epoch 400