Vision NLP Autoencoders

Autoencoders for vision and NLP.

• Vision Autoencoders: Fully connected and convolutional autoencoders, with layer-inverse constraints.
• Text Denoising Autoencoder: Sequence-to-sequence LSTM-based model for reconstructing clean text from noisy input.

Structure

vision/
    models.py       # Vision autoencoder architectures
    train.py        # Training loops for vision models
    eval.py         # Evaluation and plotting for vision
    utils.py        # Device and transform helpers

nlp/
    models.py       # NLP autoencoder and classifier architectures
    data.py         # Dataset classes, noise, vocabulary
    train.py        # Training loops for NLP models
    eval.py         # Evaluation for NLP
    utils.py        # NLP dataset stats and printing

common/
    plotting.py     # General plotting utilities
    metrics.py      # Accuracy, parameter counting
    config.py       # Centralized hyperparameters

scripts/
    run_vision.py   # Entrypoint for vision experiments
    run_nlp.py      # Entrypoint for NLP experiments
    run_transfer.py # Entrypoint for transfer learning