07L - PCA, AE, K-means, Gaussian mixture model, sparse coding, and intuitive VAE

[@ShihgianLee]

I am awake at 52:09 as well 🤣Joke aside, 2021 videos are quite different from 2020, which is a great treat! I am being introduced to VAE from EBM's R(z). Also, thanks for sharing the homework 3 questions which help me to think and understand EMB better. Thank you Professor Yann and Professor Alfredo! 🥰

[@anondoggo]

Time stamps 00:06:55 - Training methods revisited 00:08:03 - Architectural methods 00:12:00 - 1. PCA 00:18:04 - Q&A on Definitions: Labels, (un)conditional, and (un, self)supervised learning 00:25:31 - 2. Auto-encoder with Bottleneck 00:27:40 - 3. K-Means 00:34:40 - 4. Gaussian mixture model 00:41:37 - Regularized EBM 00:52:08 - Yann out of context 00:53:24 - Q&A on Norms and Posterior: when the student is thinking too far ahead 00:53:58 - 1. Unconditional regularized latent variable EBM: Sparse coding 01:06:10 - Sparse modeling on MNIST & natural patches 01:12:18 - 2. Amortized inference 01:17:02 - ISTA algorithm & RNN Encoder 01:26:56 - 3. Convolutional sparce coding 01:36:37 - 4. Video prediction: very briefly 01:39:22 - 5. VAE: an intuitive interpretation 01:48:34 - Helpful whiteboard stuff 01:52:35 - Another interpretation

[@alfcnz]

Thank you for this! I'll add the chaptering for the English and French version. Feel free to create more of these! ❤️❤️❤️

[@alfcnz]

Added!

[@cambridgebreaths3581]

Yesss..missed the notification for few hours.....I should make a model to predict the upload of your next videos:)))

[@alfcnz]

Haha, next week, it's the last week of uploading content.

[@cambridgebreaths3581]

@@alfcnz lovely. Thanks Alf

[@НиколайНовичков-е1э]

Hello, Alfredo! Thank you for video!

[@alfcnz]

Hello 👋🏻👋🏻👋🏻 You're welcome 😊😊😊

[@hamedgholami261]

thank you for this but one request though, can you share homework for these weeks as you did for the last weeks? It is a really good practice to go through. thanks again.

[@alfcnz]

There are 3 homework assignments in total.

[@hamedgholami261]

@@alfcnz Oh I see! so you mean nyu students go through 3 assignments as well? so I'm really grateful you share them with us. thank you, sir!

[@hamedgholami261]

@@alfcnz but if you mean you don't have the intent to share them publicly, can I contact you in person and provide proof of my identity as a researcher student so that you can share homework with me and be sure that I don't share them anywhere? I am just a student and my only intent is to reinforce my knowledge. anyways, if you disagree, please tell me an alternative way so that I can reinforce my knowledge myself by coding something, thank you in advance.

[@alfcnz]

3 assignments and a final project.

[@youtugeo]

In 1:12:00 Yann Lecun says that the brain doesn't do reconstruction, that it doesn't reconstruct an input from an embedding. This seems very counter intuitive to me... Why not? What are dreams then? Aren't they reconstructions of input signals (images, sounds etc.) from some sort of embeddings?

[@youtugeo]

Ok, instead of deleting the question I will write what I think the answer is.. I think he means that these embeddings should be learned through feature extraction first (learning to generate z from y with a decoder). Then, of course they could be used for reconstruction...

[@buoyrina9669]

Are there any reference textbooks for this course ?

[@alfcnz]

I'm writing it as I'm replying to you 😃😃😃

[@buoyrina9669]

@@alfcnz thanks. Looking forward to your book

[@alfcnz]

Trust me, me too, me too 😅😅😅

[@anondoggo]

1:07:41 I still can't quite figure out how Yann visualized the columns, is it by gradient ascend (grad cam I think?) thank you!!!

[@petrdvoracek670]

I think it is the actual weights W, no grad cam required. He also mentions that each tile is basis function, which confirms my assumption.

[@AdityaSanjivKanadeees]

Hi Alfredo, once gone through this online version, do you recommend watching some parts from the offline version as well. PS: Thanks a Lot for this series!!

[@alfcnz]

What is the “offline version”?

[@AdityaSanjivKanadeees]

@@alfcnz The spring 2020 version, which was recorded in a class.

[@alfcnz]

In 2020 we had other guest lectures and a few of our lessons were also different. On the website of the 2021 edition I've listed all the videos that one should study.

[@AdityaSanjivKanadeees]

@@alfcnz Thanks a lot..

[@harisumanth]

Hi, this video doesn't have thumbnail. Just letting you know. Thanks for the amazing lectures!!

[@alfcnz]

I'll do that in the morning. Thanks! Something must have gone doing with the automatic thumbnail generation.

[@alfcnz]

Okay, added from my phone. Thanks again.

[@harisumanth]

@@alfcnz welcome!

[@ahmedbahaaeldin750]

There is a sudeen jump in the prerequisites math needed , why dont you supplement us with the best sources to keep up with u and yann in term of math ?

[@alfcnz]

Which part are you facing difficulties with? Can you point out minutes:seconds?

[@sujin.lifts7]

nice

[@alfcnz]

😇😇😇