Overview¶
- Author: J. Emmanuel Johnson
- Email: jemanjohnson34@gmail.com
- Repo: github.com/jejjohnson/uncertain_gps
We will do a quick overview to show how we can account for input errors in Gaussian process regression models.
Problem Statement¶
Standard
where
Observe Noisy Estimates
Observation Means only
Posterior Predictions¶
Linearized Approximation¶
Where we take the Taylor expansion of the predictive mean and variance function. The mean function stays the same:
but the predictive variance term gets changed slightly:
with the term in red being the derivative of the predictive mean function multiplied by the variance.
Notes:
* Assumes known variance
* Assumes
egp_moment1 = jax.jfwd(posterior, args_num=(None, 0))
egp_moment2 = jax.hessian(posterior, args_num=(None, 0))
Moment-Matching¶
Mean Predictions
where:
Variance Predictions
Variational¶
Assumes we have a variational distribution function
Other Resources¶
Datasets¶
We use some toy datasets which including:
- "near square sine wave"
- The sigmoid curve
$$
$$
- Spatial IASI Data