WiTH

‘Lunch and Learn’ Speaker Covers Metadata, ML Use Cases

Machine learning (ML) can be useful for media and entertainment (M&E) organizations but they shouldn’t expect the technology to solve all their data and other challenges, according to Ashleigh Faith, the director of knowledge graph semantic search and ML and artificial intelligence (AI) at EBSCO Information Services, a provider of research databases, e-journal and e-package subscription management.

During the second MESA “Lunch and Learn” discussion, held online March 10 via its MESAverse platform and Zoom, Faith, the guest speaker, discussed ML, metadata and related trends.

Faith directs the creation, implementation and continuous development of the EBSCO cloud based data pipeline, MLAI and knowledge graph product platforms using AWS standards and Netflix OSS reference.

She leads a team of engineers, data scientists, and ML experts, and works with the sales and marketing teams in the end-to-end Google Tag Manager (GTM) (including user testing and telemetry insights) of a multilingual visual knowledge graph and recommendation engine. Faith has a proven track record of delivering on time (agile) and within budget, as well as managing a product line with a user satisfaction rating in the upper 80th percentile on the Microsoft Usability test.

“Usually trends are things that are more like fads,” she pointed out. But the trends she discussed during the second MESA “Lunch and Learn” event are “kind of bubbling up and you’ve probably heard of some of them,” she said, predicting people will “start really looking at them seriously this year.”

The first trend she addressed during the event was the issue of “software wasteland risks.”

Moving on to ML, she said: “If you are lucky enough to have a machine learning group or person, they’re going to get you the best model ever. They’re going to make it beautiful. It’s going to be like the wonderful baby bird that you want to release out into the wild. And then you will. And then that person will walk away and never come back to it again. That is a risk because a lot of these models are overfitted.”

In ML, overfitting could mean the developer made the model perfect for testing and training, which is “never good because it’s never what happens in real life,” she explained.

What she was referring to in this case, however, was overfitting the model for your use case, she said, noting that creates a challenge also. For one thing, building a ML model “takes a while to do” and there is a tendency to think it will save money and solve all of an organization’s problems, she said, adding with a laugh: “No, it won’t. It’s certainly not cheaper.” It is not an easy way out, she said.

“The smartest machines on the planet, which are quantum computers, are only as smart as a six-year-old,” she pointed out, adding: “You know what else is as smart as a six-year-old? Cats. Think about that for a second. They need a lot of help from us.”

Pattern recognition is a key technology used with ML that enables the learning process, according to Faith, who said teaching a machine is much like teaching a child.

She went on to discuss the use of ML with data in a recent project that faced architectural challenges.

One major tip she provided to those viewing the event was to make sure the people making decisions on a project understand what you’re talking about and planning to do, and make sure to put it in layman’s terms so they definitely can understand what you’re talking about.

She went on to discuss the concept of digital twins, which are virtual representations that serve as real-time digital counterparts for a physical object or process.

Concluding the discussion, she stressed how use of these technologies can help in the standardization of metadata.

Richard Atkinson, president of the Content Delivery & Security Association (CDSA), introduced Faith.

The MESAverse is MESA’s virtual work environment for its members and communities to watch live presentations, network, and step on the stage to share opinions. MESAverse is using ICVR’s RendezVu, a proprietary, interactive 3D-world application that will allow for hybrid live and virtual events and regular day-to-day interactions that go far beyond video conferencing.

To register for future weekly Thursday “Lunch and Learn” events, click here.