variational autoencoders

E40250

autoencoder architecture deep learning model generative model latent variable model probabilistic model

Variational autoencoders are a class of generative neural networks that learn probabilistic latent representations of data, enabling them to generate new, similar samples.

Statements (50)

Predicate	Object
instanceOf	autoencoder architecture → deep learning model → generative model → latent variable model → probabilistic model →
abbreviation	VAE →
appliedTo	audio → images → text → time series data →
approximate	posterior distribution over latent variables →
assume	prior distribution over latent variables →
basedOn	variational inference →
canGenerate	new data samples → similar samples to training data →
haveVariant	beta-VAEs → conditional variational autoencoders → disentangled VAEs → hierarchical VAEs → vector-quantized VAEs →
implementedWith	neural networks →
introducedBy	Diederik P. Kingma → Max Welling →
introducedInPaper	Auto-Encoding Variational Bayes →
introducedInYear	2013 →
learn	probabilistic latent representations →
model	conditional distribution of data given latent variables →
objectiveIncludes	Kullback–Leibler divergence term → reconstruction loss →
oftenUsePrior	isotropic Gaussian distribution →
optimize	evidence lower bound → variational lower bound →
relatedTo	Bayesian inference → autoencoders → generative adversarial networks →
reparameterizationTrickIntroducedBy	Diederik P. Kingma → Max Welling →
trainedWith	backpropagation → stochastic gradient descent →
typicallyUse	continuous latent variables →
use	decoder network → encoder network → latent space →
useFor	anomaly detection → data compression → image generation → missing data imputation → representation learning → semi-supervised learning →
useTechnique	reparameterization trick →

Referenced by (1)

Subject (surface form when different)	Predicate
Kullback–Leibler divergence →	usedIn