Deep Learning Lecture 3: Maximum likelihood and information
Рет қаралды 78,504
Slides available at: www.cs.ox.ac.uk/people/nando.defreitas/machinelearning/
Course taught in 2015 at the University of Oxford by Nando de Freitas with great help from Brendan Shillingford.
@mpete0273 7 жыл бұрын
Out of the entire course this is easily the most important lecture. I've watched it several times to really internalize it.
@MadcowDeity 9 жыл бұрын
Fantastic videos, I appreciate how open you're being with the coursework!
@osamamustafa6884 2 жыл бұрын
Your energy is appreciable in this whole playlist!
@paulthomann5544 9 жыл бұрын
Thank you very much for posting these lectures! So far, I find them interesting and understandable.
@CW711 9 жыл бұрын
very clear connection between least square loss, MLE and KL divergence. Thanks.
@Nestorghh 6 жыл бұрын
Great Professor! Thanks Dr De Freitas
@dbskluvu 7 жыл бұрын
This lecturer teaches more than my lecturer in 1 hour. So good. Please teach in my uni )):
@kafaayari 2 жыл бұрын
excellent teaching.
@dragonlorder 8 жыл бұрын
Better explanation than Bishop's chapter 1 : )
@omeryalcn5797 6 жыл бұрын
bishop book is bible of ml
@andrewczeizler59 8 жыл бұрын
this needs to be turned into mooc!!!
@salemameen 9 жыл бұрын
thanks
@treflir 7 жыл бұрын
Good content for beginners ! I strongly advise against your method to simulate gaussian variables though. It is a general method (the use of the inverse of the distribution function) that is really bad for gaussian variables in terms of complexity. The two best way to go are Box-Müller's method and Marsaglia's method (and from my own experience their performance are equivalent). That being said, I enjoyed this video, thank you for sharing !
@npabbisetty 6 жыл бұрын
At the end of each video, it would be great to include a "Summary for Practitioners"...that distills the theory to practice...the why and the what.
@VictorChavesVVBC 7 жыл бұрын
At the very end, how closely related is cross entropy to that KL/MLE relationship? It seems that you have it as the variable term in 1:12:08 but I'm not sure.
@abcborgess 7 жыл бұрын
nice
@robthorn3910 7 жыл бұрын
Plural of matrix is matrices.
@cognitiveinstinct2929 7 жыл бұрын
I like the video series, but the audio needs work. The constant background buzz is really distracting.
@maiiabakhova2474 9 жыл бұрын
The guy is not good with mathematics, losing logarithms, confusing probabilities and values of density funcitions. But thanks anyway.
@kikirizki4318 7 жыл бұрын
why is the value of y axis of gaussian distibution equals to p_x_given_theta ?
@myabakhova7271 7 жыл бұрын
For the probability there are must be an integral over a small interval of length delta x which is then approximated by value of function multiplied by delta x. But he ignores it because he considers ratio where delta x will be cancelled. Technically it should be at least mentioned. As mathematician I found such skipping of steps annoying.
@Ahmedkedir 7 жыл бұрын
I am wary of Mathematician who generalize about the "guy" based on one video and use argument from authority. Please check the works of the "guy" first. I admit there are minor mistakes, but this lecture was meant as revision not as main lecture.
@myabakhova7271 7 жыл бұрын
Since when mistakes are fine in a revision lecture? First time I hear about it.
@seratonewaymar1068 7 жыл бұрын
1. He's a Computer Science professor, not a mathematics prof. The fact that he's this proficient with maths is an accomplishment in itself 2. You're getting a course (that Oxford students pay around 3000-4000£ to attend) for free on KZfaq along with the slides for FREE. Don't complain 3. If you understand what his mistakes are, it shouldn't affect you in the least. You just need to code all this once and then forget about it. 4. Be positive, spread gratitude instead of your complaints Peace. Happy Learning