Editor’s note: This is the latest in the UpTech series focusing on Artificial Intelligence brought to you in partnership between YourLocalStudio.com and WRAL TechWire. Alexander Ferguson is founder and CEO of YourLocalStudio. Links to some earlier posts in the series are embedded for your convenience and information.
CARY – Welcome back to UpTech Report Series on AI. In this video and online transcript, we sit down with Alicia Klinefelter, research scientist for NVIDIA, and ask her to help define AI and the different types of AI that we often hear about.
Alicia is an expert in her field. She has a PhD in Electrical Engineering from the University of Virginia. Since joining NVIDIA, her focus has turned more towards high-performance hardware, including machine learning circuits and systems.
- Please define AI and machine learning in her terms.
In terms of kind of the recent revolution, again for me as a hardware engineer, I think of it a little bit differently than maybe someone who is more on the algorithms or kind of theoretic side. For me, I see it more as a revolution of finally having enough compute power.
Essentially, to do a lot of these complex algorithms that, you know, have been limiting this revolution for years, because I mean a lot of the algorithms in the, you know, the underlying mathematics of machine learning have been around for decades—since the 1950s. Really, what kind of revolutionized these things is finally, in the mid-2000s to late-2000s, we’ve been having this enormous progression of compute power that has enabled a lot of us to finally kinda implement these algorithms at a larger scale. So, to me, the power of AI is really in having those resources now available that weren’t there before so.
- If you had to explain AI and machine learning in very simple terms (for example: to your mother or grandmother) how would you do it?
It’s definitely, it’s the power for machines to learn at the most general level. And it’s the power of machines to learn through, kind of, iterative training, the same way that a child might learn.
That’s usually the way I explain it to my mom is, you know, when you’re first learning as a child, you’re given a lot of concrete examples of what something is. You know, you’re given a series of blocks and eventually you learn, “this is a block.” And then eventually you learn about colors, and you can identify this is a red block.
A lot of AI is like that to me, where, you know, as long as we have the data, the training data, to essentially tell a machine, you know, “this is how you can classify different categories of things,” then it can basically learn through a series of pretty simple arithmetic choices—essentially how to identify different or new things.
I think beyond that, if my mom asks how, you know, it might be a little bit more of a complicated answer. But, it’s really just the power for computers we traditionally thought were only good at doing things we told them to do to actually be more adaptive and to actually learn on their own. And I think, you know, that’s real power that we’re starting to see now, so.
- Let’s take a look at the difference between AI and machine learning.
I usually think of AI as a superset of everything, and then, really, machine learning is the implementation of how to get to this greater concept of artificial intelligence. That’s usually where I kind of differentiate the two.
Where artificial intelligence is just this super broad concept that can mean anything essentially, and then machine learning, as I mentioned, is really that specific implementation of how you do it.
- Let’s take a look at some of the other commonly used forms. Can you explain the difference between machine learning, deep learning, computer visioning, and natural language processing?
So again, I think in this case, I would consider machine learning to be the superset of all these different type of learning methods, which deep learning is kind of one of them. And, you know, deep learning kind of only having been popularized in maybe the last decade I would say—probably around with AlexNet essentially—where you basically have these kind of multilayered networks.
And which again, I think is something that was enabled more by the compute power, so in order to be deep in your learning, you need multilayer networks, and in order to have multilayer networks, you need to have a lot of complex compute power. So I would saycompared to the really simplistic models that used to define machine learning, which can encompass lots of different, other types of network typologies besides deep network or neural network typologies.
You can even just do simple linear regression, and that can be considered machine learning. So there’s a lot of different types of algorithms and network typologies that are under that machine learning umbrella. But, you know, something like deep learning or neural networks is something that’s very specific within machine learning.
And as far as I understand, the way that I would describe computer vision is really just the ability. I think of it really as image processing essentially, is the ability for a machine to look at an image, or a stream of images, and basically detect or parse something about that image, something unique. So, I kind of genericize it as being like image processing.
But, and I know that for a while that that, you know, we have a group of computer vision experts at NVIDIA, and I know for a long time before we even got interested in AI, that they had relied on these very traditional algorithms to do a lot of their image processing—what I guess I would almost call a deterministic algorithm that someone had developed mathematically. You know, if you wanted to detect whether there was a car in an image, there was a very deterministic way, to say, go through all the pixels and then identify these particular things and you’ll figure out if there’s a car.
And then kind of machine learning came through and was able to break all of their accuracy records for doing any type of image recognition or object detection. And natural language processing, you know, as far as I understand, it’s almost like a, I mean it’s a more specific application of machine learning is how I would think about it. But it’s basically parsing language essentially, almost as if a human being would. Now I’m like, it’s basically a specific model for processing language in machine learning.
This was just a taste of more UpTech interviews to come. Stay tuned as we share the full, deep-dive interviews we had with each one of our panel of experts and our upcoming episodes focused on specific topics that will transform the way you think about artificial intelligence. All this on UpTech Report’s news series on AI.