Learning a Structured Latent
While trying to implement a basic compression algorithm, you’ve managed to stumble upon a strategy for generating arbitrary input data. But to make it work, you need to tame the latent space. Remember, a latent space is a representation of data where similar inputs lie closely together in the space. For example, word embeddings and sentence embeddings from Chapter 9, Understand Text, and Chapter 11, Model Everything with Transformers, are examples of a learned latent space for text.
Unfortunately, the emphasis here is that your latent space is learned. That is to say that it’s difficult to query the latent space without access to information from the encoder. To generate images, you need access to a target image—which ...
Get Machine Learning in Elixir now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.
