Loading timeline…
20132010s
Variational Autoencoders (VAE)
A generative probabilistic model that compresses data into a latent space and reconstructs it.
Why It Was Important
Introduced by Kingma and Welling, VAEs learn a continuous 'latent' representation of data types (like faces or handwriting). By mathematically navigating this latent space, developers can generate entirely novel combinations, profoundly advancing the mathematics of generative AI alongside GANs.
Who Invented It
Diederik Kingma & Max Welling
Machine learning researchers fusing Bayesian inference and deep learning.
Applications
- Generative Media
- Anomaly Detection
- Dimensionality Reduction
Key Papers
- Auto-Encoding Variational Bayes
Diederik P. Kingma, Max Welling · ICLR 2014 · 2013
Videos
What are Autoencoders?
IBM Technology