https://github.com/davidtranhq/pytorch-rbm

A restricted Boltzmann Machine trained using Persistent Contrastive Divergence implemented with Pytorch.
https://github.com/davidtranhq/pytorch-rbm

pytorch restricted-boltzmann-machine

Last synced: 2 months ago
JSON representation

A restricted Boltzmann Machine trained using Persistent Contrastive Divergence implemented with Pytorch.

• Host: GitHub
• URL: https://github.com/davidtranhq/pytorch-rbm
• Owner: davidtranhq
• Created: 2022-05-23T06:57:32.000Z (over 3 years ago)
• Default Branch: main
• Last Pushed: 2022-05-31T18:57:45.000Z (over 3 years ago)
• Last Synced: 2025-02-02T18:47:05.862Z (8 months ago)
• Topics: pytorch, restricted-boltzmann-machine
• Language: Python
• Homepage:
• Size: 12.1 MB
• Stars: 1
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Restricted Boltzmann Machines

See the accompanying [post](https://davidtranhq.github.io/2022/05/23/generating-images-with-restricted-boltzmann-machines.html) 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:

![Generation](results/generation.png)

Images reconstructed from the model:

![Reconstruction](results/reconstruction.png)

L1 loss of the model

![Loss](results/loss.png)

Visualization of the weights in random hidden units:

![Weights](results/weights.png)