This is great content! Thanks for the valuable information that you share for those people that cannot attent the GP summer school! I would really appreciate if someone could reference some sources that elaborate on the different techinques of validating the posterior GP models. For example things like using mean square error for measuring the accuracy of the mean, or counting the new points falling inside 95% confidence interval to measure whether the confidence intervals of the posterior GP make sense. Also checking the predictive distribution by normalising the residuals.

Would have been good to see a clear presentation of posterior mean calculation and nll calculation *without* using inducing points.

Any one who can say something about the kernel k(x,x'). Here what does it mean by x and x' I thought they are like two inputs of a random variables that gives a value, but I saw something like a vector k(x,x')=x^Tx'???. Does it mean that x is the observed points and x' is point for prediction?

Arguably the best presentation on this subject.

Such a great presentation. I wish I could see the rest of the presentation.

Thank you!

I loved learning that a diagonal noise term can help with ill-conditioned matrix inversions.

Great video on causality!

If you were targeting an international audience, you should articulate and reduce your speed to make it easier to understand !

Can your model deal with exogenous control variables ? Often in literature denoted u(t)

Very good introduction to GPS

where is the slide pls🥲

I really appreciate for this. Been studying GP with lots of confusion. This was light for me.

Thanks!!!

I appreciate how thorough this video was. Many tutorials on GPs tend to handwave a lot of the mathematical details of Gaussians and use sloppy notation (which is a huge problem with machine learning education in general, in my opinion).

20:32

Its amazing. Shall I contact with this professor for some questions.

Great talk! Thank you

very very nice thanks Dr!

Amazing and insightful presentation! Thanx for publicly sharing this!

Very important topic, thanks for the talk

Woah! Great introduction!

Great talk! The first 30 minutes really helped ease in to the topic

Nice explanation! Very useful for my thesis :)

Thanks for the nice explanation.

Thanks for the nice explanation.

Starts at 4:54

Would be nice if they picked a speaker who could speak English properly. Very hard to understand (and I'm a native English language speaker)

Really good complement for the Rasmussen & Williams Gaussian Processes for Machine Learning Chapter 3 which is quite involving.

I don't see the slides on the website. the speaker says they are there ..

thank you so much for this wonderful video. In most of your figures, you have about 10 different colors that are moving along the x-axis. On each slice (a vertical line at x=x_j lets say j is 5) we have 10 points in 10 different colors that are normally distributed while they are correlated based on a kernel with the 10 points in x_i {i=0,1,.., 5, ..,n}, and lets say n is 20. My question is how do you generate this point? In total we have 10 (colors) * 20 (n) = 200 points which have to satisfy 2 conditions: 1) being normally distributed at each section 2) following the correlation based on kernal. Thank you

The boris johnson comment is gold :))))

Thank you for these classes, very helpful - and probably COVID for enabling them to go to KZbin :-)

great talk. Not easy to find structured materials on this

Cool!

Great lecture. Thank you :)

My poor english + teacher defects in speaking + strict and fast UK english = I'm not able to enjoy the lesson :(

A great introduction to GPs, it's concise and very visual. Kudos to you Dr Wilkinson! Thanks for uploading these, I hope to attend one someday :D

Hi are the slides available ?