this is the best LDA video I have seen. thank you so much.
@AlgoNudger26 күн бұрын
Thanks.
@annawilson382427 күн бұрын
10:47
@bobitsmagic4961Ай бұрын
On the slide of 33:00 we are using the Jacobian instead of the hessian. When the network only has a single output and we use the least squares loss function would the newton step collapse to gradient descent with the gradient divided by its length? It feels like we are just throwing away all curvature information at this point
@JackRid-k5sАй бұрын
Nice content just wondering why your views are so low
@AdrienLegendreАй бұрын
Excellent presentation
@matthieudegeiter3709Ай бұрын
Very nice lecture ! Thank you very much !
@mohammadhoseinrezaee-d1sАй бұрын
do anyone have the exercise of this course?
@farshidshateri_wpАй бұрын
man's mind ❌ Human mind ✅
@annawilson3824Ай бұрын
1:20:00
@annawilson3824Ай бұрын
50:55 Bayesian Inference is not hard (c)
@HerzbergTestaАй бұрын
The best course as ever... Thanks so much.
@HerzbergTestaАй бұрын
The best lecture as ever
@prateekpatel6082Ай бұрын
In the GMM ELBO , why is the q(z) in denominator missing . The ELBO looks incorrect ?
@leeris19Ай бұрын
Cool explanation and visualizations!
@ahmedhamza39392 ай бұрын
I don't understand how A is independent of C|B translated to if i told you B was can you make any statement of A and C independent of each other ?
@Amulya72 ай бұрын
Absolute goldmine
@sitrakaforler86962 ай бұрын
Title: "Foundations of Machine Learning: Walking Through Linear Regression" Introduction to basic concepts of machine learning - Course aims to prepare students for advanced machine learning courses - Focus on developing key concepts and intuitions behind machine learning Machine learning aims to detect patterns in data and make useful predictions in challenging situations. - Machine learning involves training an algorithm by giving it data and answers, allowing it to discriminate without explicit rules. - The focus of machine learning is on making useful predictions rather than learning about the world. Introduction to different types of machine learning problems - Supervised learning involves labeled data to distinguish classes - Unsupervised learning clusters data without labels, focusing on different kinds of animals Simple linear regression involves predicting a continuous variable based on one predictor. - - It uses a linear function with two parameters - intercept (beta zero) and slope (beta one) to fit the data. - - The loss function for linear regression is the mean squared error, which measures the squared deviation between actual and predicted values and is used to optimize the model. Introduction to Baby Linear Regression with a Single Parameter Beta - The model simplifies linear regression by ignoring the intercept and using only one parameter, beta. - The optimization process involves finding the minimum of a quadratic loss function using baby gradient descent with a learning rate. Understanding the challenges with non-convex functions and choosing the right learning rate in gradient descent. - Non-convex functions can lead to challenges in finding the global minimum using gradient descent. - Choosing the right learning rate is crucial as a large learning rate can cause divergence, while a small learning rate can lead to slow convergence. Explaining gradient descent for simple linear regression - Computing gradient using derivative with respect to beta not x - Utilizing derivative to update beta and converge to minimum point Understanding beta as a vector in two dimensions and its update rule using gradient - Beta can be considered as a vector with two coordinates, beta 0 and beta 1 - The gradient is a vector consisting of partial derivatives along each coordinate
@alexboche13492 ай бұрын
Great lecture thank you! At 39:24, to compute the covariance, I found his explanation incomplete because he doesn't address variation in x. I give a more detailed derivation below. cov(f(x,θ),f(x',θ)) \approx cov(f(x,θ_*) + J(x,θ_*)(θ - θ_*)), f(x',θ_*) + J(x',θ_*)(θ - θ_*)) by Taylor of J in θ = J(x,θ_*) cov((θ - θ_*)), (θ - θ_*)) J(x',θ_*)' by multilinearity of cov = J(x,θ_*) Var(θ) J(x',θ_*)' \approx J(x,θ_*) ψ^-1 J(x',θ_*)' by the Laplace approximation to posterior on θ. As for the negative, is that a typo? I thought he said it was but then he said it wasn't? I'm confused.
@sangraampatwardhan15732 ай бұрын
Yes, the negative sign was indeed a typo
@annawilson38243 ай бұрын
40:52
@sevdaebrahimi71993 ай бұрын
Thank you so much for this great course.
@saripallijitendra35733 ай бұрын
It would be nice to have the option of slides in English and captions in English if possible :)
@annawilson38243 ай бұрын
1:21:28
@shubhajitchakraborty3 ай бұрын
Would you please make videos in English? I'm from BHARAT 🇮🇳🙏🏻.
@florentin31413 ай бұрын
Is it correct that the way we define x on slide 10 the order matters? Otherwise p(x|f) would not be a probability distribution. I think this is quite inconsistent with the way x was used before: Not as some vector in {0,1}^n but as the number of glasses-wearing people -> Would be more consistent to use a binomial coefficient as normalization constant.
@jakobpcoder4 ай бұрын
Danke fürs hochladen!
@electric_sand4 ай бұрын
Tübingen ML has some of the best educational KZbin thumbnails. Usually very clean.
@AlgoNudger4 ай бұрын
Thanks.
@TharunanJR4 ай бұрын
good video
@enlightened81164 ай бұрын
Best video so far on ANOVA
@annawilson38244 ай бұрын
1:23:40
@blup7375 ай бұрын
next lecture please
@graedy26 ай бұрын
The best video on this topic I have found so far by a large margin. Excellent work!
@sumankhatri26796 ай бұрын
Hi , Please provide code and excercise of this very nice code.
@sumankhatri26796 ай бұрын
Can we get course website
@richardm59166 ай бұрын
You are the best teacher in the world thanks
@rolanddeui38436 ай бұрын
It was mentioned earlier that the product of two GP is another GP only if it is over the same set of variables (x), and that it is some else if it is over two different set of variables (say x and y). Does not this apply to the prediction step at 1:17:11 (from 2nd to 3rd line)?
@Pedritox09536 ай бұрын
Great video!
@rudeprover6 ай бұрын
Having watched quite a lot regression videos I can say confidently this is something which sums up and condenses each and every thing for a beginner to grasp linear regression smoothly(see what I did there?). Thank you so much for making this public!
@seanranieri38166 ай бұрын
27:20 Really impressive, especially the pronunciation of Kolmogorov's name.
@christophec69926 ай бұрын
have you tried with silver nano wire networks ?
@edbertkwesi49316 ай бұрын
awesome
@SiqiYinEclipse6 ай бұрын
sehr gut
@nipamghorai32176 ай бұрын
Which book do you guys follow?
@sakcee6 ай бұрын
can we have the homeward or exercises of this course?
@Elena-fh6ez7 ай бұрын
Maike hat das toll gemacht!
@bithigh83017 ай бұрын
in 36:40 sum( f_i(\theta) P_i), P has dimension 3N and f(\theta) has 207? how this multiplication is possible?
@bithigh83016 ай бұрын
Answer is also on the SMPL paper with a better notation
@bithigh83017 ай бұрын
Awesome class. But the notation in slide 11 is a bit confusing, what is the advantage of having a unity vector and rotation angle on \omega_j? And, is there a typo on slide 11 should \omega² be (\omega_j)²?