Fully Connected Autoencoder for MNIST Digits

🔍 What is an Autoencoder?

An autoencoder is a type of neural network that learns to compress (encode) input data into a lower-dimensional latent space and then reconstruct (decode) the input from this compressed representation.

It is made up of two main parts:

• Encoder: Reduces input dimensions (compression)
• Decoder: Reconstructs the original data from encoded features

🧮 Mathematical Overview

Given input data x, the encoder function f(x) maps it to a latent representation z:

z = f(x) = encoder(x)

The decoder then attempts to reconstruct x as:

x' = g(z) = decoder(z)

The model is trained by minimizing the reconstruction loss, commonly Mean Squared Error (MSE):

Loss = ||x - x'||²

---

📌 Applications of Autoencoders

• Anomaly Detection: Unusual data has high reconstruction error.
• Dimensionality Reduction: Autoencoders can serve as a non-linear alternative to PCA.
• Clustering: The latent space learned can be used for unsupervised clustering.
• Feature Extraction: Compressed latent vectors can be used as inputs for downstream models.

---

🗃️ Repository Structure

File/Folder	Description
code.ipynb	Jupyter Notebook to train and visualize the Fully Connected Autoencoder
requirements.txt	Contains all dependencies to ensure reproducibility
data/	Folder that stores MNIST dataset (downloaded automatically)

---

🚀 How to Use This Repository

Using Python Virtual Environment (Recommended for Development)

1. Create a virtual environment (Windows)

   python -m venv venv
   venv\Scripts\activate

2. Install the dependencies

   pip install -r requirements.txt

3. Run the Jupyter Notebook

   jupyter notebook code.ipynb

🧾 Requirements

• Python 3.8+
• PyTorch
• torchvision
• matplotlib
• scikit-learn
• jupyter (if running .ipynb locally)

All required libraries are listed in requirements.txt and pre-installed in the Docker image.

---

📝 Author

Shakeel Ahmed