Previous chapters focused primarily on VAEs for data where samples are often treated as independent, such as individual images. However, many datasets possess inherent structure. For instance, text data consists of ordered sequences of words, time series data exhibits temporal dependencies, and social networks or molecules are best represented as graphs. This chapter extends the VAE framework to effectively model such sequential and structured information.

You will learn to:

• Adapt VAE architectures to handle sequential data, leading to models like Recurrent VAEs (RVAEs), suitable for tasks like time series modeling.
• Integrate attention mechanisms within VAEs to capture long-range dependencies often present in sequences.
• Develop VAEs specifically for graph-structured data, enabling representation learning on networks and relational information.
• Understand applications of these VAE variants in fields such as Natural Language Processing and for analyzing dynamic systems.
• Examine the connections between VAEs designed for temporal data and established state-space models.

By the end of this chapter, you'll be equipped to apply VAEs to a broader range of complex data formats, capturing the rich dependencies they contain.