cs294a Sparse Autoencoder Lecture Part 2

Показать описание

Stanford CS294A Sparse Autoencoder and Unsupervised Feature Learning Lecture Videos

Nico Zhang

Рекомендации по теме

Комментарии

He's wearing such a nice pullover <3

davidpomerenke

This for 2011!! nearly a decade old. No wonder he looks much younger!

also creating an sparse auto encoder can be achieved easily using other methods as well!

amortalbeing

The loss function which is regular loss function + KL divergence, also missing are the derivatives of the loss function.

mantrava

It seems that the lecture note on the sparse autoencoders is different from the video. The lecture note uses KL divergence, but this is not mentioned here. Is the update scheme presented in the video doing the same? I assume using the KL regularizer, it would affect all the weights, while the scheme shown in the video will only affect the bias.

juliusctw

What is the intuition behind using the sparse constraint as opposed to fewer neurons?

clapdrix

Hello,

as I saw your graph, I was a little concerned about the perfectioness of your representation and the interpretability of the hidden layer.

Did you see the the original paper of olshausen and field about sparse coding, claiming than an overcomplete basis set really leads to derivable lower level featuremaps which are consecutive in terms of their distribution propagation.

I anyways like the idea of compressed sensing for information filtering on higher level analysis tasks.
I think it is possible to combine the outcome, using residual learning, with further layers build on the learned features (classification, convolution or noise cacelling for instance).
But correct me if I am wrong.

-M

__MGR___

thx for the great post. I have a question about the parameter updating. The Eq1 and Eq2 with respect to updating p and b are not referring to the derivative terms, why do we still use SGD on training?

anynamecanbeuse

Hello to my fellow video watchers! I had a doubt, what is that X_j ~ N(0, sigma) and why would it be difficult to learn the compress representation from it?

akshathvarugeese

in minute 32 how calculate the input that maximize the hidden unit i?

amanma

cs294a Sparse Autoencoder Lecture Part 2

cs294a Sparse Autoencoder Lecture Part 1

cs294a Sparse Autoencoder Lecture Part 2

Sparse autoencoder 01

Deep Learning(CS7015): Lec 7.5 Sparse Autoencoders

Image Compression based on Sparse AutoEncoders using Tensorflow

Sparse autoencoder: parameter study 01

Sparse auto-encoder algorithm - Method I

Deep Learning Basics: Autoencoders- Part 3

Autoencoders - EXPLAINED

Vorlesung 9.1 - Autoencoder: Einführung, Keras Functional API

Intro to Deep Learning -- L15 Autoencoders [Stat453, SS20]

03 Sparse autoencoder

Sparse autoencoder: parameter study 02

Lecture 32 Autoencoder Variants I

Sparse autoencoder

Deep Learning Basics: Autoencoders-Part 1

Deep Learning: Unsupervised Learning - Part 2

Sparse auto-encoder algorithm - Method II

Autoencoders Tutorial Part 1

Deep dictionary learning approaches for image super-resolution - Pier Luigi Dragotti, Imperial

Machine learning - Deep learning II, the Google autoencoders and dropout

Simple, Efficient and Neural Algorithms for Sparse Coding

Sparsity Autoencoders

Deep Learning for NLP - Lecture 4 - Autoencoders