/MADE_pytorch

Pytorch implementation for Masked Autoencoder Distribution Estimation - binary MNIST Dataset

Primary LanguageJupyter Notebook

MADE_pytorch

Source Paper

This code is the implementation of paper -

MADE: Masked Autoencoder for Distribution Estimation

Link to Paper

References :

Code is inspired from - https://github.com/karpathy/pytorch-made

Training Data

MNIST Dataset is used for training. Pixel values are converted to binary values before training. Image size is reduced to 20 x 20 to reduce model complexity.

Network Details

MADE defines implicit joint input image probability using the chain rule. The linear connections between layers is adjusted (masked) in a way so that each output probability is the conditional probability distribution based on all previously sampled pixel values.

Network

For binary pixel value case, sampling happens assuming the bernoulli distribution.

Training Loss

Training loss is plotted after each iteration on batch of 128 images.

Training Loss

Validation Loss

Validation loss is calculated on the valid data after each epoch.

Validation Loss

MNIST Images (Original)

orig

MNIST Images (Generated)

gen