VAEase Project

This repository contains code for experiments with various autoencoder models and synthetic datasets, as described in the associated paper.

File Overview

• models.py
  Contains the implementation of all model architectures used in the experiments, including encoders, decoders, and other neural network modules.

• sync_dataset.py
  Implements the synthetic dataset classes described in the paper. This includes various data generation methods and dataset wrappers for training and evaluation.

• Other Python Files
  All other .py files are runnable scripts. Their filenames follow the pattern:

  dataset_modelmethod.py
  

  where dataset specifies the dataset used, and modelmethod specifies the model or method applied. For example, activation_sae.py runs the Sparse Autoencoder (SAE) on the activation dataset.

Example Structure

• activation_sae.py: Runs SAE on the activation dataset.
• embedding_vae.py: Runs VAE on the embedding dataset.
• fmnist_vaep.py: Runs VAEP on the FashionMNIST dataset.

Usage

To run an experiment, execute the corresponding Python script.
For example:

python activation_sae.py

For more details on the datasets and models, please refer to the comments in models.py and sync_dataset.py.