AI-powered video curation makes learning more accessible

Michael Hickins Profile picture for user Michael Hickins July 5, 2022 Audio mode
Summary:
Students have a range of different learning styles - but adapting materials into video content can be expensive. Oracle's Michael Hickins looks into the use of machine learning technology to improve the learning experience.

female student studying using laptop and learning online © BalanceFormCreative - Shutterstock
(© BalanceFormCreative - Shutterstock)

We’ve all been faced with the dreadful prospect of watching an hours-long video of a presentation, college lecture, or corporate “fireside chat.” This trend has only been exacerbated by the COVID-19 lockdown, which has turned many of us into bleary-eyed remote workers and learners.

Corporations and colleges produce hundreds of millions of hours of long-form video every day, but as the cost of editing and producing them can be exorbitant, most of us end up getting far less value than we otherwise could.

Educational Vision Technologies (EVT) was created to address this paradox, and has developed a pair of services that provides both university students and knowledge workers with more digestible versions of long-form video content, using machine learning algorithms to split lengthy videos into short video chapters of only a few minutes each, accompanied by full transcripts and other notes.

Originally developed with disabled learners in mind, the service also benefits abled students who either cannot attend class, or who find the note-taking assistance helpful, as well as knowledge and other remote workers. Monal Parmar, founder and CEO of EVT, notes:

Studies have shown that it takes more cognitive effort to take notes while trying to listen to a lecture than it does to play chess. It doesn't make sense for students to overexert their cognitive bandwidth to write everything from the whiteboard or chalkboard. Providing notes gives students the flexibility to take as few or as many notes as work best for them.

Research from Kansas State University bears out the idea that students using “externally provided lecture notes… generally achieve more on exams than do learners who review their own notes.”

At students’ own pace

The EVT service is being used by a number of departments at University of California San Diego (UCSD), as well as by a handful of other universities and a professional training organization.

The technology allows students with disabilities to “participate in an academic lecture numerous times and at the student’s individual pace,” Joanna Boval, director of the office for students with disabilities at UCSD, says in a letter.

EVT’s Parmar says the company has raised $700,000 and currently has four full-time employees and three contractors.

The company, which built its service on Oracle Cloud Infrastructure (OCI), also participates in the Oracle for Startups program, which he says has led to a number of sales opportunities. He explains:

Many cloud accelerator programs are just called accelerators in name but do little more than provide cloud credits and some technical guidance. Oracle’s does a lot more than that.

Powered by machine learning

EVT offers two varieties of its service. EVT Bloom lets customers upload their own videos for curation, while EVT Learning Systems uses a device on-premises.

EVT Bloom uses machine learning to split recordings into short video chapters and automatically creates an interactive table of contents, a searchable speech transcript, speaker summaries, and quiz questions. The titles of the short videos are set as headings on EVT’s web platform so they can be voiced by a screen reader, allowing those with visual impairments to better navigate the video content.

Parmar says the company had to overcome a number of machine learning challenges and figure out how to make the service scale. It also had to overcome hardware bugs on the on-premises device used by EVT Learning Systems, such as overheating, as well as unreliable power supplies, networking issues, and supply chain disruption. In contrast, the software offering was more streamlined to develop. “We simply transferred many of our core algorithms to the cloud and turned them into microservices,” he says.

EVT Bloom subscription plans range in cost from $33 per hour to $60 per hour of video, billed on an annual basis. AI-curated content of one-hour lectures is available within 30 minutes, and fully corrected versions within 24 hours, according to Parmar.

The COVID-19 pandemic accelerated adoption of the technology at UCSD, but the university had already been using it to further its mission of providing equitable and inclusive learning to a diverse student population. In a statement, its Chancellor Pradeep Khosla commented:

They have helped us not only improve accessibility, but also improve student satisfaction for distance learning.

Oracle comes through

Processing for EVT Bloom occurs on OCI, with the content encrypted and stored in OCI Object Storage. Parmar explains:

We developed our machine learning microservices on OCI. Over many meetings, Oracle's engineers guided us to prototype and develop our machine learning microservice infrastructure in the Oracle Cloud. For any technical challenge, Oracle engineers have been ready to support us and regularly check in to see how things are going.

Parmar notes that cloud costs could have been a gating factor for his company, as EVT has thousands of instructional videos that get streamed from its website. But it was able to make “substantial cost savings” on video streaming because the Oracle service doesn't charge egress for the first 10TB per month.

Parmar earned a bachelor of science degree in electrical and computer engineering from UCSD in 2016 and was working toward a master’s degree with a focus on machine learning, all while working on his startup, when the pandemic hit. “Something had to give,” he says.

He plans on returning to his graduate work in 2023 or 2024 -- and could benefit from his own brainchild to finish his degree.

Loading
A grey colored placeholder image