The difference between llamas and alpacas is a great way to talk about medtech AI, Hologic VP of R&D/Innovation Mike Quick says. [Photo by linaskk via Stock.Adobe.com]
When Hologic VP of R&D/Innovation Mike Quick talks about artificial intelligence and medtech AI, he draws on his personal experience as an amateur alpaca farmer.
He and his wife had a herd of nearly a dozen alpaca when they lived in the Boston area. Then they moved to Arizona, where they now have three alpaca on a small farm in Phoenix.
“They’re a lot of fun,” he said in an interview with Medical Design & Outsourcing. “… The difference between llamas and alpacas — because it’s a common misconception of what they are — is a great way to talk about AI and the difference between machine learning and deep learning and how to learn to tell two different things apart.”
You start with the side-by-side differences between the features of a llama and an alpaca.
“Alpaca, for example, have straight ears. llamas have curved, banana-shaped ears. Llamas are about six feet tall, whereas alpaca stand about five feet tall,” he said. “Alpaca are bred for fiber to make yarn and hats and gloves, whereas llamas are bred for being pack animals that carry materials for people in the mountains. … Alpaca are sweet and cuddly, whereas a llama will listen to death metal and kill you in your sleep in terms of their personality — very different.”
Having that list of criteria is one way of thinking about machine learning, Quick said.
“We trained algorithms being able to look at the difference between the two. And that’s a very common way that we initially think about how we do training of algorithms for image recognition,” Quick said. “It’s very effective and we’ve we’ve used it in image analysis for years.”
With regards to deep learning, that’s more about taking examples of the different features or objects and classes that you’re looking for
Hologic VP of R&D/Innovation Mike Quick [Photo courtesy of Hologic]
“If you were to look at image after image labeled as llama or alpaca, you would quickly intuitively learn the features without having to call out and say, ‘Is this six feet tall or five feet tall?’ You would just recognize it as being similar or different into those two groups, and you quickly would establish what those criteria are that differentiate them,” Quick said.
“It’s a nice way to talk about the difference between machine learning — where you provide a set of features that an algorithm can can apply to an image — versus deep learning, which is providing tens of thousands of examples,” he continued. “Depending on how hard it is to tell those things apart, that is really the challenge of how we develop algorithms using deep learning and AI.”
Read more from our interview with Quick: How Hologic tapped AI and volumetric imaging for cervical cytology — and potential applications beyond