Denoising and Variational Autoencoders

A video about autoencoders, a very powerful generative model. The video includes:
Intro: (0:25)
Dimensionality reduction (3:35)
Denoising autoencoders (10:50)
Variational autoencoders (18:15)
Training autoencoders (23:36)
Github repo: www.github.com/luisguiserrano/autoencoders
Recommended videos:
Generative adversarial networks: • A Friendly Introductio...
Restricted Boltzmann machines: • Restricted Boltzmann M...
Matrix factorization: • How does Netflix recom...
Singular value decomposition: • Singular Value Decompo...
Neural networks: • A friendly introductio...
Convolutional neural networks: • A friendly introductio...
Recurrent neural networks: • A friendly introductio...
Logistic regression: • Logistic Regression an...
Shannon entropy: • Shannon Entropy and In...
Book by Luis Serrano! Grokking Machine Learning. bit.ly/grokkingML
40% discount code: serranoyt
0:00 Introduction
0:13 Generative models
3:03 Variational autoencoders
3:45 Dataset of images
10:16 Denoising autoencoders
10:27 Linear methods
10:53 A friendly introduction to deep learning and neural networks
11:58 Mapping the real numbers to the interval (0,1)
12:23 Sigmoid function
12:41 Perceptron
15:02 Correct noise
18:20 Autoencoders as generators
20:16 Latent space
23:41 Training a neural network - loss function
25:18 Training an autoencoder
25:32 Training autoencoders
25:46 Reconstruction loss (Mean squared error)
26:31 Reconstruction loss (log-loss)
27:11 Training a variational auto encoder
Correction: At 30:05, the number in the middle of the red graph should be 0.4, not 0.3.

[@khaliliskarous2225]

This is so beautifully done! What you emphasize is why the methods work, based on data-inetrnal correlation, not just how the steps of how the machines described, DAE and VAE, calculate. Thank you!!

[@shaz-z506]

Very intuitive explanation, thank you Luis for making this topic easy to understand.🙂

[@carlosgarciagomez2365]

Brilliant explanations and huge work on the video! Really appreciated.

[@ocamlmail]

Thank you so much! So clear and helpful! Really great job.

[@aparnadinesh226]

Really beautiful explanations & examples. very well explained with animations

[@damianliu1507]

Love your videos, I was just about to look into this subject!

[@DoppiaDx]

amazing explanation, thank you

[@amirrezamehrabi7240]

Brilliant explanation

[@pseudounknow5559]

As always your video is a masterpiece !

[@SerranoAcademy]

Thank you so much, glad you liked it! :)

[@woojay]

So good. Thank you.

[@mathisallyouneed6187]

Very nice explaination for beginners!

[@SerranoAcademy]

Thank you! :)

[@GuiguiGamer03]

Thanks, that was really helpful !!!

[@franciscoalmeida5016]

Great video!

[@isabelfsv]

Definitivamente el mejor!!

[@samirelzein1095]

great as usual!

[@SerranoAcademy]

Thank you Samir!

[@engineren5451]

super nice! Thanks a lot!

[@SerranoAcademy]

Thank you, glad you liked it!

[@francescserratosa3284]

🙂Excellent!

[@mostechroom9780]

Hi, great video, how do you make your amazing illustrations and animations ?

[@amirhamidi9970]

thank you for your great video. just a quick note that at time 13:08, we get the value of -1 and not 1.

[@SerranoAcademy]

Yikes, you're right! Thank you for the correction!

[@Chill_Magma]

Hello there Mr Serrano this is a great intuitive video but I have a question . On 28:26 I believe that the mean and variance may vary from input to input right? So doesn't that mean that the purple and green distributions should become dissimilar on each sample inserted? Thank you Mr Serrano.

[@fairuzshadmanishishir8171]

Make videos on Graph Neural network please

[@arvyzukai]

30:05 has a typo on the graph. q3 should be 0.4 and not 0.3 in order to sum to 1 and also to be inline with the calculations (where 0.4 is used). Just a note :) Great content by the way!

[@SerranoAcademy]

Ahh you're right, thank you so much for the correction! I can't change the video but will add a comment below.

[@xentox5016]

at 25:00 shouldnt it be a "large/small loss" instead of a "large/small loss function"? Im not really shure if i understood this right

[@andywub]

you're saving my graduate research lmao

[@udaya2008]

@SerranoAcademy Where did the first idea/intuition occurred for the first researcher who proposed that the equal diagonal values have property to denoise? you are just riding on the idea, better if you can share more information on it.

[@paedrufernando2351]

million claps from me

[@aidanthompson5053]

2:14