Developing Deep Learning models from scratch to better understand the end-to-end process. This involves:
- Creating classes for common deep learning layers.
- Creating network classes that access layers to build customizable architectures.
- Developing loss functions and optimizers from scratch to better handle issues related to training.
Reproducing this paper from scratch by manually building a GAN architecture stacking LSTM, RNN, Dense and Activation layers. Using my own library allows me to manage and monitor the end-to-end process and customize the training process so that different layers of the network are trained at different times.