This whole set of videos on machine learning is so well done and everything was explained in molecular details. Great teacher with exceptional teaching ability! I feel truly blessed.
Thank you. This lecture alone has consolidated many fragments of knowledge that I have about linear regression! It's like almost everything clicked for me. I do still have a big question. Why is the standard deviation also estimated by minimizing the log-likelihood? What makes it an appropriate estimate of the standard deviation of the same normal distribution that has the mean (x^T)*theta_ML?
Great stuff, although I wonder, should the normalisation constant for the multivariate normal pdf at 19:00 contain a factor (2*pi)^(-n/2) (since it's stated as a general multivariate Gaussian)? If it's still supposed to be the bivariate example, I missed that...
at 1:12:38 is a bit confusing. I think it should be the information that the unfair coin toss reveals to us is less than one heads-or-tails. am I missing some thing?
i'm not sure about this, but the way I undertand entropy is as a measure of randomness, thus when you have a fair coin, you have the highest entropy since all events in state space are equally likely. If you have an unfair coin you gain more information about what the value will be next time coin is flipped. If you take limiting cases you have max info gain and min entropy since every throw will result in 0 or 1. In later lectures when he talks about decision trees and information gain he explains this.
I feel bad for people trying to learn Machine Learning and don't were lucky to find this class as I was. Thanks Prof. Freitas!
This whole set of videos on machine learning is so well done and everything was explained in molecular details. Great teacher with exceptional teaching ability! I feel truly blessed.
beautifully linked the idea of maximising likelihood by illustrating the 'green line' @ 51:41
I truly appreciate these lectures. Thank you very much professor, great pacing, great structure, great content!
God bless you, professor Freitas!
I wish i watch this video earlier before the midterm.. Cool, your explanation is always amazing.. Thank you..
Very very clear explanation, I have spent a lot of time about learning probability, just now everything became clear.
really very smart professor!
Smart professor!
The lecture is great! It is really helpful. Thank you.
great intuition for MLE
Great lesson!
Thank you. This lecture alone has consolidated many fragments of knowledge that I have about linear regression! It's like almost everything clicked for me. I do still have a big question. Why is the standard deviation also estimated by minimizing the log-likelihood? What makes it an appropriate estimate of the standard deviation of the same normal distribution that has the mean (x^T)*theta_ML?
Great stuff, although I wonder, should the normalisation constant for the multivariate normal pdf at 19:00 contain a factor (2*pi)^(-n/2) (since it's stated as a general multivariate Gaussian)? If it's still supposed to be the bivariate example, I missed that...
Exactly, as you pointed out it should have negative n over 2 since it talks about n random variables
Excellent lecture ....
Superb!
+1 for your sense of humor! :) Great lecture.
awesome lecture
Perfecto
This is when got really interesting 22:02 typically, I'm given points and I am trying to learn the mu's and the sigma's
at 1:12:38 is a bit confusing. I think it should be the information that the unfair coin toss reveals to us is less than one heads-or-tails. am I missing some thing?
i'm not sure about this, but the way I undertand entropy is as a measure of randomness, thus when you have a fair coin, you have the highest entropy since all events in state space are equally likely. If you have an unfair coin you gain more information about what the value will be next time coin is flipped. If you take limiting cases you have max info gain and min entropy since every throw will result in 0 or 1. In later lectures when he talks about decision trees and information gain he explains this.
Machine learning... Linear regression
Thanks Internet for making this accessible in india.
39:00 - Maximum likelihood
45:20 - Linear regression