Videos

Transfer Learning



Siraj Raval

Transfer learning is a statistical technique that’s been getting more attention lately that enables you to reuse a model for a different task than what it was trained for. In this episode, I’m going to show you how to use transfer learning to predict instances of gold deposits using publicly available satellite imagery. We’ll discuss the 4 different mathematical ways to frame the transfer learning problem, then look into how a U-Net architecture works. The U-Net is really popular for image segmentation and classification tasks on sites like Kaggle, so knowing how it works will be a useful skill. Enjoy!

Code for this video:
https://github.com/llSourcell/mapo

Want more education? Connect with me here:
Twitter: https://twitter.com/sirajraval
Facebook: https://www.facebook.com/sirajology
instagram: https://www.instagram.com/sirajraval

Goldspot:
https://goldspot.ca/

Natural Language Processing:
https://www.youtube.com/watch?v=bDxFvr1gpSU

Convolutional Neural Networks:
https://www.youtube.com/watch?v=FTr3n7uBIuE

Watch Me Build an AI Startup:
https://www.youtube.com/watch?v=NzmoPqte4V4

Make Money with Tensorflow 2.0:
https://www.youtube.com/watch?v=WS9Nckd2kq0

Watch the first episode of my podcast:
https://www.youtube.com/watch?v=IWAe_mdTslc

Biotechnology:
https://www.youtube.com/watch?v=FTr3n7uBIuE

Join us in the Wizards Slack channel:
http://wizards.herokuapp.com/

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:
https://theschool.ai/

Signup for my newsletter for exciting updates in the field of AI:
https://goo.gl/FZzJ5w

And please support me on Patreon:
https://www.patreon.com/user?u=3191693

Source

Similar Posts

43 thoughts on “Transfer Learning
  1. Funny! When you say "geologist be like…" at 00:35, the guy you are showing is actually Maurizio D'Alema, an Italian politician who used to be in the Communist Party

  2. So it takes in one image. Then it copies the image as an input to several functions. Then the image is filtered in a different way at each function channel to output an augmented input image. Then image analysis happens on each channel. So you might get a thousands of different vision systems through 1 input image. A bit like having every different setting combinations of an app like Photoshop, maybe in several layers. That's complicated. We must decrease the combinations based somehow on external factor. Like a human eye adjusting to light and dark ? The brain changes its inputs by narrowing and widening the lens. You must do the same for which functions to employ, using all at once must be process heavy. How can input select which functions to use and not use.

  3. Sorry for the noob question, but when drawing the layers of u-net, what does the number above the layer signify? For example 1, 112, 224, 448, etc… Also is a documentation standard for such network architecture drawings?

  4. Hi Siraj. Thanks for the efforts you put in to. I am thinking for a topic for my Master's thesis. I want to come up with thesis and idea that will change the world just like google did from their master's thesis. I have six months time to think about my thesis topic. What is the correct path to go? I am interested in IoT and Machine Learning. What do you suggest me to do? Thank you again.

  5. Sir I am one of your biggest fans from Vadodara Gujurat India and would love to let you know that you have inspired many of us back here with your content and I really look up to meeting with you someday.

  6. Thanks Siraj you are not a dope thanks for this it gets really depressing. There is a specific reason for this spirit is a part of intellegence. A major part.

  7. All your videos are great specially the old ones, but the new ones are really distracting because of the moving screens you use. Please stop using those moving live screens and use some static ones like the old days. The graphs and plot and the way you use to explain problems are great but the only problem is that moving screen with bubbles and water effects are not good and make us dizzy. – – and Thanks

  8. WARNING!!!
    The safest way to avoid getting scammed in binary and Forex is by trading yourself, manage your account by yourself and monitor your profits by yourself. Deposit and Withdraw by yourself too. Stop looking for someone to help manage your account. Who said you can’t manage your account by yourself and still make profits ? Don't give out your Wallet ID and password to anyone.
    NOTE: Its not difficult but it requires consistency, determination and hard work
    Contact me via email for more help and info:binarytradegroups(at)gmail(.)com

  9. Hi Siraj, thanks a lot for this. It was back to the roots with mthods & programs and state of the art technology which I really like and where I really get valuable new input (unlike those slightly overcaffeinated "how to get a PhD in maths in 2 hours" contributions). I wonder if you could do one on CapsNets? I think those are really, really fascinating. Keep up the good work! & Regards from Germany!

  10. Awesome video, I am using transfer learning for gold detection as well and its awesome that i stumbled into your video. Gave me some great ideas!

  11. Great Siraj this video help me to understand concept of transfer learning
    But I have one doubt can we use any trained model as starting point of new model ? just as example model build on general daily objects can be used as starting point of x-ray images ? This example as images are completely different

  12. Does Transfer learning comes in Machine Learning or Deep learning? Neural networks come in Deep learning and transfer learning uses
    Neural networks..ugh I'm confused 🙆🙆🙆 please help..

  13. Such methods might be used to find important deposits on the Moon and Mars, of water, uranium, rare-earth minerals, and lithium. All of those would needed for human life and rocket fuel, power, electronics manufacture, and batteries, respectively.

Comments are closed.

WP2Social Auto Publish Powered By : XYZScripts.com