The latest technology and digital news on the web

Human-centric AI news and analysis

How to build a apparatus acquirements model in 10 minutes

I like to divide my apparatus acquirements apprenticeship into two eras:

I spent the first era acquirements how to build models with tools like scikit-learn and TensorFlow, which was hard and took forever. I spent most of that time activity afraid about all the things I didn’t know.

The second era–after I kind of knew what I was doing – I spent apprehensive why architecture apparatus learning models was so damn hard. After my crisis cleared, I took a analytical look at the apparatus acquirements tools we used today and accomplished this stuff is a lot harder .

That’s why I think the way we learn apparatus learning today is about to change.

It’s also why I’m always captivated when I ascertain a tool that makes model-building fun, intuitive, and friction-less. That couldn’t be more true of Teachable Machine. It’s a free program, build by Google, that lets you train deep acquirements models right from your browser. It’s so easy that even elementary school kids can use it!

Today, you can train image, pose, and sound models with Able Machine.

Screenshot of able machines UI

Under the hood, it uses TensorFlow.js to train a model, which is nice for a couple of reasons:

  • Training happens locally, in the browser, so none of your data gets sent the cloud. Go data privacy!
  • When training is done, Able Apparatus lets you export your model so you can use it later, in your own app, from Javascript.
  • By design, TensorFlow.js uses your GPU (if you have one) to train models, so it’s kinda speedy!

If you want the whole scoop, check out my latest adventure of #MakingWithML:

 


This commodity was accounting by Dale Markowitz, an Applied AI Engineer at Google based in Austin, Texas, where she works on applying apparatus acquirements to new fields and industries. She also likes analytic her own life problems with AI, and talks about it on YouTube.

Published October 15, 2020 — 08:27 UTC

Hottest related news