5/16/2021 Spatial Prediction | Module 5: Environmental Data and Gaussian Processes | Data Analysis: Statistical Modeling and Computation

s: Statistical Modeling and Computation in Applications | edX

MITx 6.419x Help HuitianDiao

Data Analysis: Statistical Modeling and Computation in Applications
Course Progress Dates Discussion Resources

Course / Module 5 Environmental Data and Gaussian Processes / Spatial Prediction

Previous Next
4. The Role of the Covariance Kernel
Bookmark this page

Exercises due May 21, 2021 19 59 EDT

The Role of the Covariance Kernel
These have actually a covariance
structure
with this exponential function kernel
that I had in my previous slide.
And as I said in lecture 1, this actually
looks already fairly close to being a
slice in this weather
map, potentially.
So next, we look a little bit more at
 the effect
that various forms of this kernel
function could have.

 6 03 / 6 03  1.50x    

Video Transcripts
Download video file Download SubRip (.srt) file
Download Text (.txt) file

Previously, we have presented an example of a covariance kernel defined as

2
‖𝑍 1 − 𝑍 2 ‖
𝑘 (𝑍 1 , 𝑍 2 ) = exp − .
( 2ℓ
2 )

However, the obvious questions are

Why use such a kernel function?
Can I use any function as a kernel?
Why use such kernel function?

The use of this kernel function allows for a relatively easy and computationally efficient way to parametrize the
correlations. Since the kernel function might be defined on the whole space, it allows for smooth computations
over the support of the variables to be estimated. Such a parametrization allows for the introduction of other
information like smoothness and dynamic behavior.
https://learning.edx.org/course/course-v1:MITx+6.419x+1T2021/block-v1:MITx+6.419x+1T2021+type@sequential+block@gp_lec2/block-v1:MITx+6.419x+1T2021+type@vertical+block@gp_lec2-tab4 1/3
5/16/2021 Spatial Prediction | Module 5: Environmental Data and Gaussian Processes | Data Analysis: Statistical Modeling and Computation in Applications | edX

Can I use any function as a kernel?

The short answer is no. In general, any arbitrary function whose arguments are 𝑍1 and 𝑍2 will not be a valid
covariance function.

Kernel Function
1 point possible (graded)
Definition 4.1 A function of two arguments and , mapping its inputs to is called a kernel.
𝑘 𝑍1 𝑍2 ℝ

Can a kernel such that

𝑘 be used to build a covariance function?
𝑘 (𝑥, 𝑦) ≠ 𝑘 (𝑦, 𝑥)

Yes
No

Submit You have used 0 of 1 attempt

Moreover, covariance matrices need to be positive semi-definite. Thus, one needs to guarantee that the kernel
function's output creates a positive definite matrix or a square of matrices.
Definition 4.2 If the covariance function is translation invariant, then it is called stationary.
This will happen if the covariance function depends on . 𝑍1 − 𝑍2

Definition 4.3 If the covariance function depends only on a norm |𝑍 1 − 𝑍 2 | , then it is called isotropic.
This means that the covariance function depends only on the distance between and . 𝑍1 𝑍2

Definition 4.4 If the covariance function depends on , then it is called dot product covariance.
𝑍
⊺

1
𝑍2

Kernel Function
1 point possible (graded)
Is the the following kernel function isotropic?

7.6
2
‖𝑍 1 − 𝑍 2 ‖
𝑘 (𝑍 1 , 𝑍 2 ) = exp − ,
( 2 )
2ℓ

Yes
No

Submit You have used 0 of 1 attempt

Discussion Hide Discussion

Topic: Module 5 Environmental Data and Gaussian Processes:Spatial Prediction / 4.
The Role of the Covariance Kernel
https://learning.edx.org/course/course-v1:MITx+6.419x+1T2021/block-v1:MITx+6.419x+1T2021+type@sequential+block@gp_lec2/block-v1:MITx+6.419x+1T2021+type@vertical+block@gp_lec2-tab4 2/3
5/16/2021 Spatial Prediction | Module 5: Environmental Data and Gaussian Processes | Data Analysis: Statistical Modeling and Computation in Applications | edX

Add a Post

Show all posts by recent activity

There are no posts in this topic yet.


Previous Next

All Rights Reserved

edX
About
Affiliates
edX for Business
Open edX
Careers
News
Legal
Terms of Service & Honor Code
Privacy Policy
Accessibility Policy
Trademark Policy
Sitemap
Connect
Blog
Contact Us
Help Center
Media Kit
Donate

© 2021 edX Inc. All rights reserved.

深圳市恒宇博科技有限公司 粤ICP备17044299号 2

https://learning.edx.org/course/course-v1:MITx+6.419x+1T2021/block-v1:MITx+6.419x+1T2021+type@sequential+block@gp_lec2/block-v1:MITx+6.419x+1T2021+type@vertical+block@gp_lec2-tab4 3/3