Machine, Deep, or Edge Learning? What’s the difference? | Cognex AI for Factory Automation
Machine Learning? Deep Learning? Edge Learning? What do all these terms mean and how are they used in the context of machine vision and factory automation?
Watch as Reto Wyss walks us through these concepts and explains the power behind example-based artificial intelligence.
Learn more at: https://www.cognex.com/blogs/deep-lea...
Introduction to Deep Learning for Factory Automation: https://www.cognex.com/what-is/deep-l...
#ai #deeplearning #edgelearning #machinelearning #machinevision
TRANSCRIPT:
Reto Wyss, VP, AI Technology, Cognex:
Machine learning and deep learning, they're really technical terms. And if you want, deep learning is actually a subset of what machine learning is. Machine learning is the general theory of how can machines do tasks by learning from samples.
If we talk about "machine vision", typically what we mean is "rule based" solution. An engineer sits down, looks at a problem, thinks of some simple rules or complex rules that can solve that problem and then basically programs an algorithm in order to do that. If he wants to find a red circle in an image, he has to sort of define a rule that sort of says: well first of all find those red pixels, and then I have a rule that decides whether those pixels are really aligned on a circle. And if all of that holds true, then I say, okay, now this is my circle, right? So these are the rule based approaches.
In machine learning, we flip everything around and say, well actually we don't want to define those rules up front, but rather we want to define examples that define what kind of rule set are we looking for and then basically this system learns based on those examples in order to solve the task. Instead of having an engineer who sort of decides what kind of rules need to be implemented and tested -- bottom up sort of -- we have a learning-based approach where the task is basically left to the algorithm to find out what is the proper configuration of those rules in order to come from that input image to a specific decision or output.
You need a relatively large amount of images in order just to get started. So the compute you really need to both train but also to execute your model is quite demanding. Those two things in combination make it actually fairly complex. So with edge learning what we're trying to do is really sort of tackle this by developing a technology which allows you to learn from much fewer images much quicker. And so much quicker that actually it's not only that you don't need a GPU, you can actually even do it on an embedded device in a matter of less than a second.
--
ABOUT US:
The world’s leading provider of vision systems, software, sensors, and industrial barcode readers used in manufacturing automation. Cognex vision helps companies improve product quality, eliminate production errors, lower manufacturing costs, and exceed consumer expectations for high quality products at an affordable price.
HELPFUL LINKS:
Support: https://www.cognex.com/en/support
Resource Center: https://www.cognex.com/resources
Blogs: https://www.cognex.com/blogs
LinkedIn: / cognex-corporation