 |
| The use cases that Inception members are working on. |
There are currently 4,200 startups in the programme, 24% in the Asia Pacific region (APAC) and a further 5% in India. More than a third, 37%, are in the Europe, Middle East and Africa region (EMEA).
Over the past 12 months, NVIDIA has strengthened the ecosystem with a network of Inception incubators and accelerators sround the world, and is organising Go-to-Market Connect events somewhere i the world every two weeks.
Lamonde was speaking at an Inception showcase for an audience of venture capitalists, during which selected startups introduced themselves:
BabbleLabs
BabbleLabs' Clear Cloud technology is enhancing speech quality, accuracy, and personalisation with a combination of neural networks, digital signal processing (DSP), and speech science.
Dr Chris Rowen, CEO, said that the deep learning-based technology “works across languages, in almost any noise background”. He also shared that BabbleLabs Clear Command has 10x better accuracy than general cloud-based speech recognition.
Explore:
Watch before-and-after videos of BabbleLabs' speech enhancement at work onYouTube
Mapillary
Mapillary's solution is to use third-party imagery of the same location to generate 3D maps using AI. To date, Mapillary has 480 million images and 37 billion objects indexed and detected globally.
“All of this is available,” Solem said, for mapping, automotive and smart cities applications. “We're building a living representation of the world that can be used for many applications.”
PRENAV
PRENAV aims to solve infrastructure maintenance problems. According to Nathan Schuett, CEO, a lot of hours are spent on manual inspections and maintenance and yet infrastructure can still fail, from bridges to dams.
While drones have been highlighted as a solution, GPS-guided drones cannot get close to infrastructure, whereas people manually operating drones might not examine a crucial object or make the right judgment, Schuett said. PRENAV's answer is a fully-automated AI-powered drone whose path is first mapped, and which can get close to the infrastructure to inspect it all and produce high-quality imagery.
“We're training algorithms to find common damage and defects,” he said. “These are customised for each industry vertical.”
Lighthouse customers for PRENAV's retail and leased systems as well as as-a-service offering include organisations which maintain bridges and which need inspections in confined spaces.
“Our vision is to digitise the world's infrastructure,” Schuett said. “You have to be right up close.”
Shone
Ugo Vollmer, CEO of Shone, thinking big. Shone is going beyond autonomous cars and trucks to create autonomous cargo ships. Existing ships are retrofitted with Shone's autonomous technology, which improves safety, reduces crew size and enhances fuel efficiency.
Unlike cars and trucks, which can leverage the Internet, ship technology has to run locally. Use cases include hazard detection, working 24x7 based on data gathered from the ship's sensors, and fatigue reduction.
Lilt
 |
| Lilt's translation system depends on a human to correct its translation in order to learn. |
Lilt is all about neural machine translation. Instead of having AI do all of the translation, a human translator works together with the AI to come up with a translation, a process called 'human-in-the-loop'. This partnership helps the AI improve as well.
Spence Green, Lilt CEO, noted that high-quality translation has become better, faster and cheaper, but it is still expensive, so companies tend to select what they translate into other languages. He estimates that 400 words in English would take a human translator 1.5 hours to translate into another language, at a cost of US$80 per language.
Neural machine translation technology from Lilt can lead to a 70% reduction in human translation errors, a 3-5x increase in human translation speed, and a 60% reduction in human translation costs, Green said.
“Lilt is the first translation automation system with human feedback,” he said. “It alters the economics of enterprise translation.”
VyasaAnalytics
Vyasa Cortex is a deep learning-based data science platform that was originally targeted at life sciences and healthcare, but which has since branched into other sectors. Mike Hugo, CTO, showed how data such as text and images can be analysed to arrive at desired outcomes.
Kinetica
 |
| Farouk lists some of Kinetica's customers, including the Lippo Group and Telkomsel. |
Kinetica is an evangelist about making your data work for you. Instead of looking at data as a passive asset, it can be integrated into a data lake to yield insights, says Paul Appleby, CEO of Kinetica.
The company built a cloud-ready, GPU-accelerated platform for active analytics that allows developers to build new classes of applications.
Irina Farouk, Chief Product Officer, Kinetica, shared that how a US energy exploration company, Anadarko, used to take days to analyse 9 mllion data points. With Kinetica, the company can now analyse 90 billion high-fidelity data points instantly, giving more assurance that predictions are accurate. Risk calculations, typically an overnight batch process, can be calculated in real time with Kinetica, Farouk said.
GOAT
 |
| Shin explains how the shoes in an image are actually identified as being the shoes actually listed for sale. |
Andy Shin, CTO of the world's largest sneaker marketplace, is fighting fraud with AI. Sneakers are big business, which has generated a whole industry making counterfeit sneakers, he explained. “There is a one in 10 chance that if you buy any shoe online it's likely to be fake,” he said.
When the platform was smaller, tamping down on fraud manually was easy. Now that GOAT has 150,000 sellers and 12 million buyers, GOAT ensures that the sneakers on sale on its platform are genuine with a combination of machine learning and human expertise.
“There are now so many users that we can't verify manually,” he said.
One scam is listing a shoe before it is actually available, as it is possible to ask for much more money for it. Three deep learning models check if the image in the listing is even a shoe; then if there are shoes in the image, if they are the actual ones listed for sale. Another inference is whether the seller actually has the shoe by checking if the seller has copied the image from elsewhere.
“Machine learning is the only way to scale handle classic marketplace problems,” Shin said.
Hashtag: #GTC19
*NVIDIA sponsored transport and accommodation for GTC.
No comments:
Post a Comment