Video

If A and B nodes have two edges to each other, in the adjacency matrix (UNDIRECTED G) is it Aij=1 or 2(there's an edge yes=1, there are two edges =2)? Which one is correct? Also if A-B there's no edge but B-A a has an edge(direction) it should be calculated for both right?

Do we have a book which teaches just like this lecture series

how is dconcept = d x V different from the U x ⨊ Matrix if both give the position of the User to Rating in the new coordinate space

2:56 - RL 3:26 - why RL is difficult

Gemini 1.5 Pro: This video is about a mixture of experts model, a neural network architecture proposed by Geoffrey Hinton in the early 1990s. The main idea is to train a set of neural networks, each specializing in a different part of the data. A manager neural network then assigns weights to these experts to decide which expert to rely on for a particular case. The benefit of this approach is that it allows the models to specialize on a subset of the data, leading to better performance on that subset. This is particularly useful for very large datasets. However, training a mixture of experts model is more complex than training a single model. Here are the key points of the video: * Mixture of experts model consists of multiple neural networks (experts) and a manager neural network. * Each expert is trained on a different part of the data. * The manager neural network assigns weights to the experts based on the input data. * This allows experts to specialize on a subset of the data and improve performance. * Mixture of experts model is particularly useful for very large datasets. * Training a mixture of experts model is more complex than training a single model. I hope this summary is helpful!

This was great to watch! Thank you and God bless!

missed u , prev prof was horrible

this video is the best to SVD, the best, the best!!!

for the last slide, you should impose the condition in the minimization that the sum of y is 0. otherwise the constant vector of all 1's is a trivial soln (corresponding to lambda_1)

such great presentation with great insights. so we can interpret a vector as a scalar function on the vertex set. multiplication with the adj matrix can be seen as averaging out this function wrt to its neighbors. so in a way, this is like the linear-algebraic version of the heat equation

best explanation of PCA 🔥🔥

this is too boring and the more sad part is that it comes from Stanford uni

Thank you for this video, it has been very helpful

Turkish subtitles please ❤

我谢谢你啊

I LOVE JURE!

very clear and straight forward, thank you for making this video

Insightful!!

At 4:25 I don't understand why there is only one block of size 8. It looks like there should be two of 8 and the block of size 16 should be shifted by 8 (and then the same reasoning could apply). However he computes everything as if it was ok, am I missing something?

When a user does not learn mic placement 101 when recording video, a bad user experience will be the result. Isn't it ironic

So what's the difference between negative loss likelihood and cross-entropy?

Great introduction.

Very well explained, thank you so much!

Hes genius

They are saying that the decomposition is unique, but this isn't true.

good !

5:40 those are even human can't understand.

I never really understood the concept behind SVD until now. The example in the first minute made everything click!

Helped me totally grok! Thanks so much prof.

Great lessons, thank you Stanford.

I wonder if liquid/ODE echo state nets could model high-dimensional systems better

This is legendary introduction of RMSProp to wider world, paper was only written after this. Interesting enough almost on the next slide where Geoffrey recommends algorithms to pick he suggests RMSProp with Momentum which is indeed the principle of Adam - current industry standard.

Thanks. I was able to understand the concept of SVD. Very clean explained

How do I calculate that fiedler vector?

when exactly we passed from x being just a vector to supposing that x is an eigenvector? Thanks in advance

@5:35, Please correct your statement. SVD is applicable for both Real and Complex matrices.

Prof is excellent in delivering essential pieces, I find this very informative

How you describe this study should realistically prove the corrupt person's in authority > though what should correlate are those people that are intent on totally abusing all our fundamental rights which includes the corrupt Malta government staff, lawfirms, Notaries, Police, judges who use Bogus Court Claims to steal all our assets

And when you don't want to be connected, but are being intellectually trafficked, through frequencies you help promote? like myself, at this time? You don't care, until you are held a silent hostage, ignored, trafficked, silenced.

the notation for factoring R as Q and P can be made better as follows : Q ----> matrix of shape (k x m) where m = no. of movies and k = latent vector dimensions P -----> matrix of shape (k x n) where n = no. of users and k = latent vector dimensions R ~ transpose(Q)*P this gives a nice view as the latent vectors are all packed as columns in both matrices P and Q

indians are the best teachers

Thank you for your clear explanation and great effort sir. And what is evaluation techniques to evaluate these content based recommendation systems sir, Could you please answer it will be more helpful sir

Thank you very much!!!

Godfather of AI ❤🙏

Hinton is one of the greats in machine learning so how can he mess up the indices this badly?

Amazing video. All clear now.

I would love to see a little explanation of the complexity that this creates, and how it is better than pairwise comparison. How is e.g. the pairwise comparison of band hashes avoided in this circumstance?

Is the equation at the slide 22 correct? I mean, the SVD there is not the approximation, right? The optimal error equation only makes sense when applied over the truncated SVD approximation. Otherwise, that subtraction would be equal zero.

bla bla bla ¨!!!!

He was in the wave earlier than gpt