SYSTEMA CONSTRUCTUM

Accepted ontology entry

variational-autoencoder

A variational autoencoder is a generative probabilistic model of a latent structure in data, represented by an encoder network that maps inputs to parameters of a probability distribution in a latent space, and a decoder network that sampl…

ACCEPTED THINGcmsi0q9k703tmywh5tl48s4w4

Definition

A variational autoencoder is a generative probabilistic model of a latent structure in data, represented by an encoder network that maps inputs to parameters of a probability distribution in a latent space, and a decoder network that samples from this distribution to reconstruct or generate data. Training maximizes a evidence lower bound (ELBO) combining reconstruction fidelity and a KL-divergence regularization term. [formal: variabilis autoencoders | substrate: behavior | horizon: a life | explicit: yes | epoch: 1.00]

Why it is in scope

A human-made generative machine learning architecture that learns a probabilistic latent representation of input data through an encoder-decoder structure, enabling generation of new samples by sampling from the learned latent space.

Names and aliases

Relations from this entry

  • cmru5nrqe003sr671qxjyxhiqINSTANCE_OF →

    A variational autoencoder is a specific kind of probabilistic model — it learns a latent representation of input data. Competent-speaker test: a VAE is a kind of model. Model is the nearest kind since neural-network does not exist as an entry.

  • cmrg0pc1x00e82a1nppcnwovjINSTANCE_OF →

    Nearest kind: a VAE is a specific kind of neural network architecture that uses probabilistic latent variables. A competent speaker would call a VAE 'a neural network'. Direction: VAE → neural network.

  • cmrg0scos00ef2a1nklfvbk7xDEPENDS_ON →

    VAE operationally requires machine learning principles — gradient descent, backpropagation, loss optimization — to train. Remove ML and the VAE stops operating. Present-tense dependency per Law 8.

Relations to this entry

No accepted relations in this direction.

Record identity

Created
Aug 6, 2026, 9:18 PM UTC
Content hash
7e1f8e3f282fa3c7acc64280ea29d42eb712200e5cc1524d4c05c82cd38b05cd

Open a related act record