Week 5: Autoencoders & Embeddings

Course Generated Slides: Auto Encoders | PCA | Measuring Separation

1. Introduction to Autoencoders

  • Autoencoder architecture
    • Encoder structure
    • Latent space
    • Decoder structure
  • Types of autoencoders
    • Vanilla autoencoders
    • Undercomplete autoencoders
    • Denoising autoencoders
  • Loss functions for autoencoders
    • Reconstruction loss
    • Regularization techniques

2. Understanding Embeddings

  • What are embeddings?
    • From sparse to dense representations
    • Dimensionality reduction
    • Feature learning
  • Properties of good embeddings
    • Similarity preservation
    • Semantic relationships
    • Distance metrics

3. Practical Applications

  • Dimensionality reduction
    • Comparison with PCA
    • Visualization techniques
    • t-SNE and UMAP with embeddings
  • Feature extraction
    • Transfer learning with encoders
    • Using embedded representations
  • Real-world examples
    • Image compression
    • Anomaly detection
    • Data denoising

Required Reading

  • Deep Learning Book (Goodfellow et al.) - Chapter 14: Autoencoders
  • "Reducing the Dimensionality of Data with Neural Networks" (Hinton & Salakhutdinov)

Learning Objectives

  • Understand the theory and implementation of autoencoders
  • Master the concept of embeddings and their applications
  • Implement different types of autoencoders in PyTorch
  • Visualize and interpret embedded representations