Lecture 3 | Machine Learning (Stanford)

[sonicblare]

At 32:How is that the 32:30, the professor says that the error is gaussian, then writes down on the other board saying , this implies that the probability of p( y/x;theta). I don't understand this implication. Can someone point me to a link or tell me how to get to the next step?

[JordanSoet]

Does anyone know why when he changes to the likelihood definition from the probability, the y gets a little arrow on top of it?

[wpeng001]

@JordanSoet because y is a m dimensional vector

[avico510]

wow he makes p values and error terms so complicated. the concepts are more important than the proofs for most

[harrycook111]

I think the question asked around 25min is really good. Really, the non-param method is like a purely interpolation method while param. method builds a model that helps extrapolate some output, whether correct or not depends on the system to be modeled.

[astroboomboy]

Anybody have any suggestions for me as to what kind of math I need to study in order to understand this? I have bought a book in calculus, but the book was very hard to learn from, but I will start putting in all my time come christmas. What kind of other math books should I get?

[MultiYolgezer]

@astroboomboy

mit ocw classes single and multivariable calculus and linear algebra will suffice both have free lectures on youtube

[zhenjiezhang]

you need a good linear algebra text book.

[royrdrgz]

Proffesor Andrew, these lectures are the highway to get into the machine learning word!!

[axeld93]

Stanford students are great! They aren't afraid to ask questions and challenge the assumptions made in the analysis. They are independent thinkers! This would never happen in my university.

[Jacob011]

@axeld93 Huh, my case exactly! People at out university do just the opposite of Stanford's. Nobody knows what the heck is going on and everybody is afraid to ask so as not to look as a dumass. :D

[akytable]

Pr. Ng mentioned that the discussion sections were to be recorded. Are they online?

[alexplanation]

Wow, Professor Ng does a lot of work to say that function isn't gaussian, which is mysterious because it's totally gaussian! He calls it a gaussian by accident like five times, too!

[thundercookies1981]

@alexplanation his point is that the interpretation of the function in that context shouldn't be construed as having anything to do with probability distributions. The function just happens to be a convenient one for LWR; as he said, you can use others, and some do.

[VancouverData]

in this lecture he covers loess regression, maximum likelihood, and logistic regression

[TheCrappyaccount]

great lecture series. The mathematics is very nice and I can see it will have many applications. Thank You.

[amyzylzl]

good explanation. Thank you.

[kunchichekkan]

If you can't understand the math, that is not the professor's problem.

[phamnamlong]

2:17 : should it be theta transposed times x superscripted (i) instead of x without superscript ?

[prateekmi2]

@phamnamlong Yes, it should be.

[ceokevin]

show what all Galileo and others did, how people started with neural nets and that didn't quite work and why... Show the applications in ads prediction, handwriting recognition and hundreds of other fields. The guy can't even answer questions! "We'll talk about it in a later lecture" sure. I follow the maths and stats just fine but there's no need to show off, Stanford!

[darfunkelidas]

@ceokevin Well you can always make your own and post them on youtube. :)

[ceokevin]

This isn't a great lecture, no not at all. He had great attendance in his first lecture and almost everyone's gone because he's made it so maths-heavy. Machine learning is a fun topic and though it's good to know the underlying maths that drives the algos, it would be better to first present the concepts, talk about them at a high level, then later delve into the maths. Explain why they work or make sense, give some historical background because there is a LOT of history ...

[FrankCashio]

@ceokevin

These lectures are great. I implemented the liner regression model that he talked about last lecture in a afternoon because he showed the 'oh so scary' math.

[idunnononame]

@ceokevin I love these lectures. I've been trying to understand this material from wikipedia and other things I could find online. If you find the mathematics difficult, know it's much harder without him laying it all out for you.

[circusboy90210]

wow robot learning & porn music together again :P

[skarootz]

It looks so interesant...

[gekorio]

Does anyone knows if the lecture notes are available on the internet? It would be perfect do not need to pause the video every time to take them.

[sayhitotim]

@gekorio

[sayhitotim]

@gekorio try searching for "andrew ng machine learning" and look for his homepage. Find CS229, the projects are listed at the bottom of the page

[natapolsri]

You should have a description that give an outline content for each lecture.

[gekorio]

he has lol

[seborinos]

@natapolsri it does just look at the description of the video for the links =D

[utuber420]

He shd visit Pakistan to be endowed with a probabilistic semantics of 0.

[leonoel]

HE is already doing it easy XD

[david04268]

Thank you so much for sharing this great lecture!

[mosaguitar]

it is not working the video no longer avaliable

[zhaoyangster]

it is even harder while your mind was not there ^-^

[Wolfnoriil]

It's called abstracting. We use abstraction because we humans are not capable of coping with 1000s of low level concepts at once, duh.

[Adovid]

@Wolfnoriil wrong, its because there is a difference between understanding a concept in terms of an abstraction vs only being able to understand something in terms of the basic components. If you only see a car as an arrangement of parts you will not be able to infer that it could be useful as a mode of travel.