Like two valedictorians, SimInsights and Photomath have a story worth hearing about how AI is advancing education.
sim insight Located in Irvine, Calif., uses NVIDIA conversational AI to bring lifelike virtual and augmented reality lessons to college students and employee training.
Photomass — Founded in Zagreb, Croatia and based in San Mateo, California — Uses computer vision and natural language processing to create apps to help students and their parents brush up on everything from math to calculus. did.
Both companies NVIDIA Inceptionis a free global program that fosters cutting-edge startups.
surfing sims in california
Rajesh Jha has loved simulation ever since he developed a physics simulation engine for mechanical parts in college over 25 years ago. “So I put Sim That’s the name I started my company with in 2009,” he said.
SimInsights originally developed web and mobile training simulations. Jha secured a grant to develop his HyperSkill when AR and VR platforms became available. Now the company’s flagship product, cloud-based, is his AI-powered 3D simulation authoring and analysis tool that makes training immersive.
The software helped UCLA’s medical center build a virtual clinic to train students. However, they were unhappy with the low accuracy of their rule-based conversational AI, so Jha took data from the first class and trained his network with Deep His Neurals using . NVIDIA RiverGPU-accelerated software for building speech AI applications.
Riva powers voice AI
“The quality improved quickly. They say it’s the most realistic workout they’ve ever used,” says Jha.
Now UCLA hopes to apply this technology to train thousands of nurses to deal with infectious diseases.
“Conversational AI will play a big role in education and training because it personalizes the experience,” he said. “And a lot of research shows that people learn more and retain it longer if they can do that.”
Access to new technology
SimInsights is an NVIDIA Inception member, so early access to Riva and NVIDIA Taois a toolkit that uses transfer learning to accelerate the evaluation and training of AI models. They have become a standard part of the company’s workflow.
Regarding Riva, “It’s powerful software and our team is very grateful to be working with NVIDIA to brainstorm the next steps,” said Jha.
Specifically, SimInsights aims to develop large-scale conversational AI models with more features such as question answering so that students can point to objects in the scene and ask questions about them. .
“Riva gives us more capabilities, so we’re building them into HyperSkill to make digital learning as good as working with an expert. It’s going to take a while, but this is there It’s a way to get there,” he said.
Accelerating Mathematics in Croatia
In Zagreb, Damir Sabol got stuck trying to help his eldest son figure out a homework math problem. The idea for Photomath, an app that has been downloaded over 300 million times since its release in 2015, was born.
This app detects equations from photos on your smartphone and presents step-by-step solutions in a format that supports different learning styles.
Ivan Jurin, who leads the startup’s AI project, said:
Some teachers have students open apps instead of working on the blackboard. This is an anecdote that colors Julin’s day.
“We want to make education more accessible,” he said. “The free version of Photomath can help people without resources understand math in much the same way as those who can afford a tutor.”
Large hybrid model
Internally, one large neural network does most of the work, detecting and solving equations. It is a combination of a convolutional network and a transformation model packed with about 100 million parameters.
Trained on local server NVIDIA RTX A6000 GPUsFor a cost-sensitive start-up, “Training in the cloud has not motivated us to experiment with large datasets or more complex models, but using a local server allows us to experiment when necessary. can be queued by the company’s engineers.
Once trained, the service will run in the cloud. NVIDIA T4 Tensor Core GPUhe described it as “extremely cost-effective”.
Inference Acceleration with NVIDIA Software
The startup is moving to a full stack of NVIDIA AI software to accelerate inference.it contains NVIDIA Triton Inference Server To maximize throughput, TensorRT software development kit to minimize latency and NVIDIA Dalia library for fast processing of images.
“We used the open source TorchServe, but it was not as efficient as we had hoped,” says Vekić. NVIDIA software “uses it on a small model to achieve 100% GPU utilization and converts a large model to that as well.”
This is a technical challenge that NVIDIA experts can handle, and one of the benefits of attending Inception.
SimInsights and Photomath are among NVIDIA Inception’s 10,000-plus combined members of hundreds of startups making education smarter with machine learning.
Check out the next GTC session for more details. NVIDIA River, NVIDIA Tao When NVIDIA Triton and TensorRT.
