Restricted Boltzmann Machines

Implementation Details

• The implemented RBM consist of 2 layers with 784 visible units and 256 hidden units.
• Trained using K-Contrastive Divergence with k = 1.
• The RBM can be modified as per the use.
• It is built solely from scratch using NumPy.
• It is trained for MNIST Reconstruction and learning its probability distribution.
• Trained on 60000 images.
• It is modular in nature so you could change the parameters as per your choice.

How to Use

• Download the train_images from here
• For sample demonstration a clipped image is reconstructed using trained RBM.
• The notebook is ready to run.

Generative Adversarial Networks

Variational Autoencoders

For learning

• To learn the concepts please watch the videos
• The notations are also similar.

Have Fun! Learning 😃