Like two valedictorians, SimInsights and Photomath tell stories worth hearing about how AI is advancing education.
SimInsights in Irvine, CA uses NVIDIA conversational AI to make virtual and augmented reality courses realistic for students and employee training.
Photomath – founded in Zagreb, Croatia and based in San Mateo, Calif. – has created an app using computer vision and natural language processing to help students and their parents learn everything from arithmetic to calculus.
Both companies are part of NVIDIA Inception, a free global program that supports cutting-edge startups.
Surf Sims in California
Rajesh Jha loved simulations ever since he developed a physics simulation engine for mechanical parts in college over 25 years ago. “So I put sim in the name when I started my own company in 2009,” he said.
SimInsights initially developed web and mobile training simulations. When AR and VR platforms became available, Jha secured a grant to develop HyperSkill. Now the company’s main product, it is a cloud-based, AI-powered 3D simulation creation and analysis tool that makes training immersive.
The software helped the UCLA Medical Center build a virtual clinic to train students. But they complained about the low accuracy of its rule-based conversational AI, so Jha took the data from the first class and trained a deep neural network using NVIDIA Riva, a GPU-accelerated software to create voice AI applications.
Riva revolutionizes voice AI
“There’s been a rapid increase in quality, and they say it’s the most realistic formation they’ve used,” Jha said.
Today, UCLA wants to apply the technology to train thousands of nurses in infectious disease management.
“Conversational AI plays a huge role in education and training because it personalizes the experience,” he said. “And lots of research shows that if you can do that, people learn more and retain it longer.”
Access to new technologies
Since SimInsights is a member of NVIDIA Inception, it gained early access to Riva and NVIDIA TAO, a toolkit that accelerates AI model evaluation and training through transfer learning. They have become standard parts of the business workflow.
As for Riva, “it’s powerful software, and our team really appreciates working with NVIDIA to think about our next steps,” Jha said.
Specifically, SimInsights aims to develop broader conversational AI models with more functions, such as answering questions so that students can point to objects in a scene and ask questions about them.
“As Riva gives us more capabilities, we will integrate them into HyperSkill to make digital learning as good as working with an expert – it will take some time, but this is the way to get there,” he said. -he declares.
Speeding up math in Croatia
In Zagreb, Damir Sabol got stuck trying to help his oldest son figure out a math problem in his homework. This sparked the idea for Photomath, an app that has been downloaded over 300 million times since its release in 2015.
The app detects an equation in a smartphone image, then displays step-by-step solutions in formats that support different learning styles.
“At peak times, we get thousands of requests per second, so we have to be very fast,” said Ivan Jurin, who leads the startup’s AI projects.
Some teachers ask students to open the app as an alternative to working on the blackboard. It’s the kind of anecdote that makes Jurin’s day.
“We want to make education more accessible,” he said. “The free version of Photomath can help resource-poor people understand math almost as well as someone who can afford a tutor.”
A great hybrid model
Under the hood, a large neural network does most of the work, detecting and analyzing the equations. It is a mixture of a convolutional network and a transformer model which contains around 100 million parameters.
It is trained on local servers with NVIDIA RTX A6000 GPUs. For a cost-conscious startup, “training in the cloud hasn’t motivated us to experiment with larger datasets and more complex models, but with local servers we can queue up experiments like we see fit,” said Vedran Vekić, a machine learning manager. engineer in the company.
Once trained, the service runs in the cloud on NVIDIA T4 Tensor Core GPUs, which it described as “very cost effective”.
NVIDIA software accelerates inference
The startup is migrating to a full stack of NVIDIA AI software to accelerate inference. It includes NVIDIA Triton Inference Server for maximum throughput, TensorRT SDK to minimize latency, and NVIDIA DALI, a library for fast image processing.
“We were using open source TorchServe, but it wasn’t as efficient as we hoped,” Vekić said. The NVIDIA software “gets 100% GPU utilization, so we use it on our smaller models and convert our large model to it as well.”
It’s a technical challenge that NVIDIA experts can help solve, one of the benefits of being in Inception.
SimInsights and Photomath are among hundreds of startups – out of a total of more than 10,000 NVIDIA Inception members – making education smarter through machine learning.
For more, check out these GTC sessions on NVIDIA Riva, NVIDIA Tao, and NVIDIA Triton and TensorRT.