Automatic Machine Learning

Automatic Machine Learning or "AutoML" is a field of Artificial Intelligence thats gaining a lot of interest lately. The idea is that doing any kind of task related to machine learning involves a whole lot of steps like cleaning a dataset, choosing a model, deciding what the right configurations of that model should be, deciding what the most relevant features are etc. The goal of AutoML is to automate all of that up to a point where all a data scientist would need to do is tell a machine to perform some task using a dataset and wait for it to learn how by itself. In this episode, i'm going to explain several popular AutoML techniques, then compare top AutoML frameworks like AutoKeras, Auto Sklearn, h20, Ludwig, etc. to help you decide which one will be the best for your needs.

Want more education? Connect with me here:

AutoML Framework #1 - MLBox

AutoML Framework #2 - Auto Sklean

AutoML Framework #3 - TPOT

AutoML Framework #4 - H20

AutoML Framework #5 - Autokeras

AutoML Framework #6 - Ludwig

My video on Ludwig:

The DARTS paper:

DARTS in PyTorch:

Cool write-up on Simulated Annealing:

Cooler write-up on Bayesian Optimization:

----------------------------------------------------------------------------------------------------

Make Money with Tensorflow 2.0:

Build a Finance Startup:

Build a Healthcare Startup:

Join us in the Wizards Slack channel:

Hit the Join button above to sign up to become a member of my channel for access to exclusive live streams!

Join us at the School of AI:

Signup for my newsletter for exciting updates in the field of AI:

And please support me on Patreon:

[NutHoofd]

Hey Siraj, just wanted to let you know that I think that you've made a lot of progress in the quality of your videos and the explanations you're giving. I have not been able to work on AI for a while due to a very busy year, and the difference from a year back and now in terms of the way you speak, the words you say, the timing and clarity of visualizations (and how they focus on what you're talking about better so they add to the explanation and don't distract) have all greatly improved. Thank you very much for all the informative content you put out. Glad to see you still going, keep it up!

[MarshallRichards]

But most people don't eat from the hands of chefs everyday. They cook for themselves.

[shresthaditya]

2:03-How to frame DataScience Problem
2:48-DARTs
4:00-Grid Search and Random Search
4:30-MetaHeurstics
5:10-Simulating Annealing
7:42-ML Box
8:20-Auto-Sklearn

[connor-shorten]

Great video! I don't think the problem is that AutoML doesn't work for unsupervised learning, rather just that it hasn't been applied much. You could easily imagine using an NAS technique or DARTS on an auto-encoder or a rotation prediction task. Thanks for mentioning Particle Swarm Optimization, I was looking for a term for that! I'm curious what the differences are between Population-Based Training and Particle Swarm Optimization, or if they refer to the same idea?

[Eu_Sunt_Dracul]

Siraj you make the best of the explanations dude, awesome

[harishchoudapurkar]

Wonderful as usual!! I like the way you compress so much information in a short video.

[Malzebiear]

Hey Siraj, I agree with everything you said here, and I'm glad you are always bring up the most current topics. However, I haven't seen you make a video about the latest and best technologies for data pooling, and data manipulation. Where and how should I pool data from many data sources so that I can easily run ML models on it (including AutoML)?

[subschallenge-nhxp]

Siraj 2016-2019 he was data scientists






Automl yes

[gokulakrishnan]

I attend School of AI Program at Chennai and now I love you more than before coz you don't stop in YouTube and Take offline meetups and conference too

[bakermotosports]

Any suggestion of one of these or another that you can import JavaScript not Python? Also for pattern recognition learning. Any help would be appreciated. Great video.

[laythfadhala]

Sometimes I press like before watching your videos.
That was very interesting topic, thank you. 👍

[jaydeepvipradas]

Data itself is a kind of inference. We are supposed to find more inferences. I.e. Combining basic level inferences using And, Or etc, at multiple levels can help reach manageable set of inferences. Levels of inferences: level 1 inference is data, level 2 inference is what data is pointing to, 3rd level inference is combining inferences using And, Or, then grouping similar inferences. And then continuing the process.

[RH-jcww]

In the explanation of the Bayesian, I think the thing that you call it activation function is actually called "acquisition" function. But I'm not sure whether people call it also "activation function" or not!

[eventhatsme]

What is the difference between H2O AutoML and Driverless? Are any of these free?

[ronaldronald]

Aha. I like that! I have some ideas on ml implementations but the knoledge required to get started is a steep learningcurve. That energy is put to better use if i spend that time defining the problem and its datasets.

[MohanKumar-gjth]

Hi Siraj thanks for the awesome contents you offer very very valuable useful and inspirational. I have a quick question in a GAN can the generator be considered as a creative aspect(left brain ) and discriminators (right brain)

[othmenfarah]

Hey Siraj can u make a video on 'sparse representation and dictionary learning algorithms"

[socomajor]

Thank you for this video. My knowledge was limited to a few automl solutions I had heard about. Now I feel like I have a better understanding of what steps the automl ecosystem tries to automate, and some good candidates to benchmark for my problem.
Super useful video.

[PieroMolino]

Than you again for the shout-out to Ludwig :)

[MikeAirforce]

Thanks Siraj! Working on a paper in this area and this was a great overview!