So we build iteratively a distribution over all the functions that pass through yi. The distribution is based on Gaussian Processes and we have a formula to build it iteratively. Gaussian Processes are defined by a mean() and covariance() functions, so it is possible to calculate the "mean" of our distribution (over functions that go through yi) which is the most likely function that passes through yi.
There are some mathematical issues when you want to define GPR properly (f.e. Why are Gaussian processes determined by their mean and cov functions or what we mean when talking about “measurement over functions”) but you got the idea!
Thank you for the perfect explanation. Can we call the last part as bayesian optimization, i.e. the combination of gaussian process and conditional probability mechanism?
Absolutely no reason why we must trust the observed points 100%, right? But the regressor seems to pass through them with 0 variance which seems very wrong.
2:36 Sorry but what most people, including you, fail to explain is HOW DO N-DIMENSIONAL NORMAL DISTRIBUTIONS LEAD TO GAUßIAN PROCESSES? Are the (infinite) Gaußian distributions added on top of each other on the x-axis, on the y-axis? How can we understand this visually? This is such an abstract concept that not even this "introduction" manages to explain THE SLIGHTEST. This is not an introduction, it's just another blabla to trick the viewer into understanding a concept that can't be understood if it isn't explained better. To get more customers.
The best intro to GP ever
I agree
Thanks for summarizing it all in such a short introduction. Grateful!
Thank you! This is a really clear and understandable introduction to GP regression.
非常感谢，讲解知识简单易懂，大师
Concise and Crisp Explanation! Really liked it
Thank you very much for your clear and concise explanation.
Very good introduction - thanks!
So we build iteratively a distribution over all the functions that pass through yi. The distribution is based on Gaussian Processes and we have a formula to build it iteratively. Gaussian Processes are defined by a mean() and covariance() functions, so it is possible to calculate the "mean" of our distribution (over functions that go through yi) which is the most likely function that passes through yi.
There are some mathematical issues when you want to define GPR properly (f.e. Why are Gaussian processes determined by their mean and cov functions or what we mean when talking about “measurement over functions”) but you got the idea!
incredible explanation, thank you
Thanks for posting this great introduction.
Very nicely explained, thank you 🎉
hello,a perfect video.can you share you code?"I want to continue following your approach and proceed further."
Thanks. Very helpful.
Thank you for the perfect explanation. Can we call the last part as bayesian optimization, i.e. the combination of gaussian process and conditional probability mechanism?
Definetly!
Thank you , this is great bro
nice explanation👍
Great explanation bro !
excellent, Thank you.
this is really good intro. is there a continuation of this?
There will be … soon 🤓
Good video.
So good!
Absolutely no reason why we must trust the observed points 100%, right? But the regressor seems to pass through them with 0 variance which seems very wrong.
Awesome 💯
How do you come up with these animations? They are very good!
This is the manim Python library - you can create amazing animations with this! 🤩
2:36 Sorry but what most people, including you, fail to explain is HOW DO N-DIMENSIONAL NORMAL DISTRIBUTIONS LEAD TO GAUßIAN PROCESSES? Are the (infinite) Gaußian distributions added on top of each other on the x-axis, on the y-axis? How can we understand this visually? This is such an abstract concept that not even this "introduction" manages to explain THE SLIGHTEST. This is not an introduction, it's just another blabla to trick the viewer into understanding a concept that can't be understood if it isn't explained better. To get more customers.
Areyou from north europe?
German accent.
Video style looks copy of 3blue1brown channel🤣🤣
The library he build to do such animations is just great. We could not resist using it 🚀