latent space

latent space (plural latent spaces)

1. (machine learning) A multidimensional spatial representation of latent variables, where related concepts are positioned closer to one another.

   Synonym: embedding space

   • 2021, Valliappa Lakshmanan, Martin Görner, Ryan Gillard, Practical Machine Learning for Computer Vision, O'Reilly, →ISBN:

     If we truly want to use the decoder of an autoencoder-type structure to generate images, then we need to develop a way to organize or regularize the latent space.

   • 2022 September 13, Devin Coldewey, “A terrifying AI-generated woman is lurking in the abyss of latent space”, in TechCrunch‎, retrieved 2023-03-31:

     A chilling discovery by an AI researcher finds that the “latent space” comprising a deep learning model’s memory is haunted by least one horrifying figure — a bloody-faced woman now known as “Loab.”

   • 2023, Roshani Raut et al., editors, Generative Adversarial Networks and Deep Learning: Theory and Applications, CRC Press, →ISBN:

     As can be observed, the central idea of anomaly detection using GAN revolves around learning the latent space representation of data samples, which is then used to compute reconstruction losses to arrive at an anomaly score.

• latent space on Wikipedia.Wikipedia