Meta Learning

Meta learning describes the concept of 'learning to learn'. What if we could have AI learn how to optimize itself? An AI could learn the optimal hyper-parameters, architecture, and even dataset! Its a really interesting topic, and in this video I'll describe some meta learning techniques and focus on one in particular; deep neuro-evolution. We'll build an image classifier using a deep neuro-evolutionary algorithm. Enjoy!

Code for this video:

Please Subscribe! And like. And comment. That's what keeps me going.

Want more education? Connect with me here:

More learning resources:

Join us in the Wizards Slack channel:

School of AI:

And please support me on Patreon:

Signup for my newsletter for exciting updates in the field of AI:
Hiring? Need a Job? See our job board!:

Need help on a project? See our consulting group:

[slightlygruff]

the visuals are getting better and better. Amazing

[johnhammer]

The only click that i am sure of not regretting later is the click on your channel videos.You are amazing bro.

[bghack]

Thanks for finally adopting my ideas into practice, THANK YOU.

[rameshmaddali]

How can I say thanks to you Siraj. Love you bro for giving this kind of well worthy knowledge. God bless you. Long live my dear.

[Nibszagger]

If AI is god then Siraj is the messenger

[DheerajBhaskar]

Ends with "I have to go find a gradient". So much meta in this 😀

Because he's an IA (intelligent agent) optimised by evolutionary processes

[odingames]

After watching the first 8 seconds of this video I was admitted to a hospital.

[rfyorfyo]

neural networks are the future, the true definition of intelligence

[wolfisraging]

def to_learn(data):
return to_learn(data)

[academicconnorshorten]

Great video, interesting topic and papers!

[ruiwang]

Earlier this year, Fei-Fei Li published a paper without using reinforcement learning or evolutionary computation on meta learning and achieved state-of-the art. Their technique only need one fifth of the RL methods, not too sure about EC though.

[GurpreetSingh-thdi]

sir, can you make a detailed video about transpose convolution a.k.a deconvolution

[larryteslaspacexboringlawr]

thank you for evolutionary algorithm video

[willykitheka]

This is very very good stuff! Bite-sized and full of useful information. Thanks alot Suraj for the effort!🙌👍

[JLK]

Evolving AI's is definitely the way forward; but we have to be careful because this is also how you get to skynet...

[vulturebeast]

This is literally awesome than traditional neural networks strategy. Loved this video thanks for this great information. :)) Cheers

[tiffanyl]

Hello,
Question 1 : Is Selective Search neural evolution algorithm ? Regarding at the example it seems very similar in purpose

Question 2 : What the Environnement change mean in the context of classification for exemple ?
I can guess it in a reinforcement learning context where the environnement could be the ground like sand, rock, ...

Question 3 : On the exemple there are mutated member an non-mutated member, what mutation means regarding of those members, is it a change of value (like the value of a variable) or the add of an instruction (like a if condition) ?

Thanks

[froyorex]

Wow Siraj, loved the infographics it make learning easer. (Where are the memes ?loved those too.)

[SuperJg]

Hey Siraj, It would be awesome if you could read the book Clever Algorithms and then explain those algorithms. It is an amazing book and you would be happier, I promise.

[jh]

So the NAS(Neural Architecture Searching) is meta-learning? What about one shot learning? Im kinda confused of those concepts..