4. VAE

Activity: VAE Implementation

In this exercise, you will implement and evaluate a Variational Autoencoder (VAE) on the MNIST or Fashion MNIST dataset. The goal is to understand the architecture, training process, and performance of VAEs.

Instructions

1. Data Preparation:

   • Load the MNIST/Fashion MNIST dataset;
   • Normalize the images to the range [0, 1];
   • Split the dataset into training and validation sets.

2. Model Implementation:

   • Define the VAE architecture, including the encoder and decoder networks;
   • Implement the reparameterization trick.

3. Training:

   • Train the VAE on the MNIST/Fashion MNIST dataset;
   • Monitor the loss and generate reconstructions during training.

4. Evaluation:

   • Evaluate the VAE's performance on the validation set;
   • Generate new samples from the learned latent space.

5. Visualization:

   • Visualize original and reconstructed images;
   • Visualize the latent space (in case of a latent space until 3-D, otherwise use a reduced visualization, e.g., using t-SNE, UMAP or PCA).

6. Report:

   • Summarize your findings, including challenges faced and insights gained;
   • Include visualizations of reconstructions and latent space.

7. Extra Credit (Optional):

   • Experiment the same dataset with a Autoencoder (AE) and compare the results with the VAE;
   • Experiment with different latent space dimensions and report the effects on reconstruction quality and sample generation.

Important Notes:

• The deliverable must be submitted in the format specified: GitHub Pages. No other formats will be accepted. - there exists a template for the course that you can use to create your GitHub Pages - template;

• There is a strict policy against plagiarism. Any form of plagiarism will result in a zero grade for the activity and may lead to further disciplinary actions as per the university's academic integrity policies;

• The deadline for each activity is not extended, and it is expected that you complete them within the timeframe provided in the course schedule - NO EXCEPTIONS will be made for late submissions.

• AI Collaboration is allowed, but each student MUST UNDERSTAND and be able to explain all parts of the code and analysis submitted. Any use of AI tools must be properly cited in your report. ORAL EXAMS may require you to explain your work in detail.

• All deliverables for individual activities should be submitted through the course platform insper.blackboard.com.

Grade Criteria: