Solution of deep learning projects.
Firstly, we work on scalar backpropagation only by numpy and math. And we train on a small dataset - synth dataset to show the inefficiency.
Then, we implement Tensor backpropagation with MNIST dataset to do the image classification.
Furthermore, we work on custom Automatic differentiation to fully understand how pytorch works.
Finally, we use pytorch to implement CNNs to do image classification.
We explore the long-term dependency modelling capabilities of Recurrent Neural Networks (RNNs), Long Short-Term Networks (LSTMs) with Autoregressive. Additionally, we train a character-level language model on natural language text and experiment with character-by-character text generation.
We work generative modelling with the variational inference and GAN on the MNIST dataset/Imagenette dataset. We then use variational autoencoders and GAN on the same task and discuss how the intractability issues are resolved. We explore the performance of variational autoencoder and GAN by sampling images from the models.