See the accompanying post for this model.
A restricted Boltzmann machine implemented with PyTorch. Trained with persistent contrastive divergence, momentum, and L2 weight decay.
models/ contains the model parameters and the hyperparameters for a model trained on the MNIST dataset. The loss and generated examples for each model is pictured below.
Images generated from the model:
Images reconstructed from the model:
L1 loss of the model
Visualization of the weights in random hidden units:
