Fichas
GPs and Unsupervised Learning
Fichas del conjunto (6)
- 1Why do we use Gaussian processes?To get a prior for dual linear regression so we have uncertainty
- 2Why are GPs useful?Each instantiation is Gaussian so probability mass never goes to 0.
Infinitinely many values so no functions with zero probability. - 3What is a GP?Generalisation of a distribution so that one instantiation of the process is a distribution.
It is evaluated by the kernel covariance function - 4What is unsupervised learning?Learn a representation X from data Y instead of learning a relationship f from data pairs
- 5What are latent variables?Variables inferred from observed data
- 6How to do unsupervised learning?Specify a distribution over the latent coordinates where x is the latent space and z the inputs, p(x|,W,z) and p(z)
Intractable to reach posterior of both os do Type II Likelihood, marginialise over x and maximise w
Información
- Descripción
- Machine Learning Fichas sobre GPs and Unsupervised Learning, creado por Philip Masters el 03/01/2020.
- Creado por
- Philip Masters
- Visitas
- 1
- Valoración