Gradient Descent - THE MATH YOU SHOULD KNOW

[@emrek1]

Derivative of log(x) is 1/(ln10*x) . Since ln10 is a constant, it doesn't spoil gradient descent.

[@patite3103]

You've done an amazing work! I love your videos! I wish you could do some more math videos! For example about neural network backpropagation, perceptron convergence theorem, mixture gaussian models....At 8:39 shouldn't we add the negativ sign to the NLL function?

[@DarshanSenTheComposer]

Having watched a couple of Prof. Andrew Ng's ML videos on Coursera, watching this video is pure satisfaction. I really liked the fact that you walked through the vectorized approaches as well. Isn't ML just a glorified form of Numerical Methods? Thank you. :)

[@northsand]

This video saved my life

[@kingkong792]

love your content, thank you for taking the time to share your skills!

[@danielrosas2240]

I like these series of videos. Thank u so much!

[@saminchowdhury7995]

this is amazing brother. u made me understand maths ur a G.

[@mmm777ization]

prob of data inc. loss dec @4:20

[@dvdsct]

dude, AMAZING video, thanks!

[@kennethleung4487]

Fantastic work! Keep it up!

[@forcedlevy]

Thank you Tobias Funke

[@forcedlevy]

11:46 should'nt the update term of gradient descent be "thetha=theta-alpha*gradient" ? you seem to use addition instead of subtraction

[@kestonsmith1354]

Alpha*gradient will give you a negative number so that's why he puts the addition.

[@NajmehMousavi]

amazing and clear explanation, great work :)

[@imed_rahmani]

Thank you so much, sir. I really appreciate your videos

[@amarnathjagatap2339]

Sir make more videos on machine learning

[@95Bloulou]

Thank you ! Great video in my opinion !

[@rahuldeora5815]

I had a question which has been troubling me for quite some time. When we take the derivative of the weight matrix times the output of its pervious layer we get a transposed matrix. I never understood how the transpose appeared. I searched alot but could not find a very clear explanation anywhere. How you guide me towards where I can find it explained simply?

[@ripsirwin1]

It kinda blows my mind that the loss function is independent of the model. Is this correct?

[@danishnawaz7869]

Challenging you to make me understand the maths of gradient boosting algorithm. 😝

[@CodeEmporium]

Challenge accepted

[@ephraimwork351]

@@CodeEmporium how did the challenge go

[@sooryaprakash6390]

@@ephraimwork351 kzbin.info/www/bejne/g3qznH5rj6amo9U&ab_channel=CodeEmporium

[@yulinliu850]

Thanks a lot!

[@jose4877]

You said to add a negative sign to the l function, and then proceeded to leave functional expression exactly the same. I do not see that you added the negative sign.

[@KoKi-nx1xo]

why is the Theta transposed??

[@animeshsharma7332]

Just to make the two matrices compatible with each other for multiplication. Theta is of order (1xd), also X is of (1xd). Transposing theta will change its order to (dx1), which will make it multipliable with X (1xd).

[@moisessoto5061]

thanks alot

[@kestonsmith1354]

👌

[@CodeEmporium]

🙂