Latent Variable
Definition
A latent variable is a quantity not directly observed but inferred in a model from observable data.
Latent variables can represent hidden factors, classes, or states. Their interpretation depends on the model and is not automatically unique.
Simply explained
Like an invisible cause estimated from multiple observed clues.
Why it matters in practice
Relevant for factor models, mixture models, VAEs, and probabilistic models.
Example from business
A model describes customer behavior via a latent factor that is not directly measured.
Limits and pitfalls
- Interpret latent variable as a real physical entity; ignore identifiability and model assumptions.
Origin & history
Latent variables are a long-standing concept in statistics and psychometrics and central in modern generative models.
Use in mid-sized companies
- Possible application in SMEs: Model hidden patterns when direct measurements are missing and assumptions are clearly documented.
Frequently asked questions
Is latent the same as unknown?
It is not directly observed but modeled. Is the meaning unambiguous? Often not.
Sources
- Deep Learning, Kapitel 16, Abschnitte 16.5–16.7 · Goodfellow et al., Deep Learning, Kapitel 16, Abschnitte 16.5–16.7
Sources checked on 12.10.2026.
Editorially reviewed on: