What is an RBM (Restricted Boltzmann Machine)?

Подписаться 812 тыс.

Просмотров 33 тыс.

50% 1

Learn more about WatsonX: ibm.biz/BdPuC6
Learn more about AI → ibm.biz/what-is-ai
Check out IBM Watson → ibm.biz/Check-Out-Watson
How do those "you may also like" lists get generated? Well, a great way to do that is by using a restricted Boltzmann machine (RBM). But what actually is an RBM?
In this video, Martin Keen will answer that question and explain more about how they work and what else an RBM is good for.
Download a free AI ebook → ibm.biz/free-ai-book
Read about the Journey to AI → ibm.biz/blog-journey-to-ai
Get started for free on IBM Cloud → ibm.biz/start-free-cloud
Subscribe to see more videos like this in the future → ibm.biz/subscribe-now
#AI #Software #ITModernization #watsonx

Развлечения

Опубликовано:

5 авг 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 39

@martinfunkquist5342 Год назад

What is the difference between an RBM and a regular feed-forward network? They seem quite similar to me.

@kristoferkrus 11 месяцев назад

An RBM sends signals both "forwards" and "backwards" during inference, uses contrastive divergence for learning the weights and does not involve a loss function, while a feedforward network only sends signals forwards during inference and uses backpropagation and gradient descent for learning the weights, which requires a loss function. Besides, the RBM is energy-based (hence it has an energy function which can be said to be instead of a loss function) and follows (a simplified version of) the Boltzmann distribution (that doesn't include k and T), so it is stochastic, while a feedforward network isn't energy-based, but is instead deterministic.

@samcoding 3 месяца назад

Someone correct me if I'm wrong, but in a simpler, higher level view to what @kristoferkrus said, feedforward networks just take an input, pass to hidden layer(s) and produce an output. In that order and direction. RBMs take an input and pass to the hidden layers. Then the hidden layers pass it back to the input layers to generate the output.

@ahmedsowdagar9034 2 года назад

I was constantly searching for the examples of what is visible layer and hidden layer. This video explained me what it is. Thanks

@nemeziz_prime 2 года назад

It'd be great if IBM could make a dedicated deep learning playlist consisting of videos such as this

@vgreddysaragada 10 месяцев назад

You made it simple..elegant presentation..Great work..Thank you..

@pavanpandya9080 Год назад

Beautifully Explained. Thank you for the Video!

@amishajain3400 Год назад

Beautifully explained Thank You!

@erikslorenz 2 года назад

I am incredibly motivated to build a light board

@oualda12 2 года назад

Thanks for the video, I'am new in this domain, I want to ask if the RBM have only two layers (visible and hidden) how can we get the output of this RBM, should we add an output layer to get result or what? thank you again.

@Tumbledweeb Год назад

I have indeed made a decision; The one that brought me to this video here! I'm here looking for photos of any of these Boltzmann Machines.

@vishnupv2008 2 года назад

In which Neural network is nodes on a given layer is connected to other nodes in the same layer?

@siddharthagrawal8300 Год назад

This just sounds like a neural network without any output?

@tanishasethi7363 2 года назад

i love how he's smiling throughout the vid

@bzqp2 2 года назад

Wait. So are the weights summed up to activate the nodes in the hidden layer or does the sum represent the probability of activating of a node?

@Programmer_Cookbook 2 года назад

Both. The weights and biases are used to estimate the hidden units sampling p(h|v) or to estimate the visible units sampling p(v|h). So although we can speak about activation of units it is not a deterministic process. And in the other hand the weights and biases, along with hidden and visible units, are used to calculate the energy of the system, which is considered as probability as well.

@INSIDERSTUDIO 8 месяцев назад

No one notice but man writing in reverse 😮 how hard he train to do that 🔥🙌

@IBMTechnology 7 месяцев назад

See ibm.biz/write-backwards for the backstory

@Abhilashaisgood Месяц назад

i think they mirrored the videoo, its really cool write smthing in a glass then open your selfie cameraa , from back camera they look reversed and form front camera they lookk samee!

@mathewssaiji5149 2 года назад

waiting for your recommendation system and explainable recommendation system videos

@PunmasterSTP 2 месяца назад

Restricted Boltzmann Machine? More like "Really cool network that's just the thing!" 👍

@abdulsaboor2168 3 месяца назад

How its different from simple ann with backprotogation??

@freespam9236 Год назад

watching "AI Essentials" playlist - no recommendations engine in the play right now

@smallstep9827 2 года назад

sample?

@high_fly_bird 2 года назад

Charismatic speaker! But I think the theme of hidden layers is not clear enough - hidden layers usually are not interpreted. Maybe u were talking about hidden layers? And it woulb be cool if you actually make an example of WHAT exactly is passed to the visible layer. Numbers? Which numbers?

@hitarthpanchal1479 2 года назад

How is it different from standard ANNs

@MartinKeen 2 года назад

Thanks for watching Hitarth. Basically the thing that makes an RBM different to a standard artificial neural network is the RBM has connections that go both forward and backwards (the feed forward pass and feed backward pass) which makes an RBM very adept at adjusted weighting and bias based on observed data.

@andreaabeliano4482 2 года назад

At very high level, one main difference is that ANNs typically are classifiers, they need labels, also to train and get the weights of the edges.

@apostolismoschopoulos1876 2 года назад

@@andreaabeliano4482 using RBMs, we are not interested on the weights of the edges? Aren't the final weights the probabilities that after someone watches video A will watch video B? Am I understanding this correctly?

@Programmer_Cookbook 2 года назад

@@apostolismoschopoulos1876 Similar to ANN, In RBM we're *TOTALLY* interested on adjusting the weights and biases. And yes, a trained net (weights, biases) will tell us about the probabilities of visible units after sampling.

@Programmer_Cookbook 2 года назад

As someone told, the main difference is that ANN is supervised learning with targets to predict, while RBM is an unsupervised learning method. Other differences: Objective: ANN -> learns a complex function, RBM -> learns a probabilty function What does: ANN -> predicts output, RBM -> estimate probable group of variables (visible and latent) Training algorithm: ANN -> backpropagation, RBM -> contrastive divergence Basic principle: ANN -> decreases a cost function, RBM -> decreases an energy function (probability function) Weights and biases: ANN -> deterministic activation of units, RBM -> stochastic activation of units

@quocanhnguyen7275 Год назад

So bad, doesn't say anything. THis is just any neural network

@svanvoor 7 месяцев назад

Either (a) this guy is very good at mirror writing, or (b) they mirrored the video after recording. Given he's writing with his left hand, and given the .9 probability of right-handedness as a Bayesian prior, I assume P(b)>P(a).

@alyashour5861 9 месяцев назад

How is bro writing backwards perfectly the whole time

@IBMTechnology 8 месяцев назад

See ibm.biz/write-backwards

@danielebarnabo43 2 года назад

What the hell? Is this what you do when you're not brewing?

@flor.7797 10 месяцев назад

😂

@zzador Месяц назад

You not really knowing what you talking about. The weights are just weights and NOT probabilities. The summed activity of a unit fed through the sigmoid function is the activation probability of a unit in an RBM and definetely NOT the weights.

@Tom-qz8xw 4 месяца назад

Terrible explanation, whats the point of making these videos if you dont show equations, you didnt mention KL divergence or anything technical, practically useless