Autoencoder

In this assignment you will build an autoencoder using PyTorch for unsupervised representation learning.

You will cover:

• Encoder: Design a network that compresses input data into a compact latent vector \(z\).
• Decoder: Design a symmetric network that reconstructs the original input from \(z\).
• Reconstruction loss: Train the model end-to-end by minimizing the difference between the input and its reconstruction (MSE or binary cross-entropy).
• Latent space visualization: Reduce the latent vectors to 2D (e.g., with t-SNE or PCA) and visualize the learned representation.
• Denoising autoencoder (optional): Corrupt inputs with noise during training and evaluate whether the model learns a more robust representation.

Problems will be released closer to the due date.