Each year, Nvidia sifts through hundreds of startups that use its graphics processing units (GPUs) to accelerate non-graphics applications, such as artificial intelligence or pattern recognition. And it comes up with a handful that are worth investment or attention via its Inception program.

I’ve been to Nvidia’s GPU Technology Conference every year, but I won’t be able to make it to next week’s event because it is overlapping with the Game Developers Conference in San Francisco. But I did have a chance to collect some info on the startups in Inception, Nvidia’s virtual accelerator program that helps GPU-based startups get to market faster.

Through the program, Nvidia provides GPU discounts, technical guidance, and business networking opportunities at places such as the Inception Showcase at GTC.

This year’s batch includes:

Shone

Above: Shone is automating cargo ships.

Image Credit: Shone

Based in San Francisco, Shone is bringing autonomous cargo ships to market. It’s working with the cargo shipping industry to provide benefits similar to those offered by autonomous vehicle features in the automotive industry. Shone applies GPUs to cargo ship data, such as sonar, radar, GPS, and AIS (a ship-to-ship tracking system). This enables speedy processing of terabytes of training data on Shone’s custom algorithms to develop perception, navigation, and control systems for ocean freighters.

Maritime shipping is the backbone of the global economy, but it is not sustainable. Shone retrofits cargo ships with autonomous technologies to increase safety and efficiencies, building on top of existing sensors onboard cargo ships (radar, AIS, GPS, Anemometer, and others). It also adds a few additional sensors, primarily cameras.

Shone runs computer vision on the cameras and performs sensor fusion between the various perception sensors to build a probabilistic understanding of the ship’s surroundings. Then it highlights only the relevant information that is useful for the crew, based on the context. Shone’s smart copilot also recommends safe and efficient decisions to the crew.

Goat

Above: Goat wants you to trust its authentic sneaker marketplace.

Image Credit: Goat

Goat Group raised $100 million in new funding for its sneaker marketplace that uses AI to authenticate shoe listings for sought-after collectible kicks. The Culver City, California startup has expanded to more than 500 employees, attracts more than 10 million users, and has been growing at an astronomical rate since its founding in 2015.

Its authentication process leverages machine learning and human expertise to bring trust and transparency to the sneaker industry. Through its innovative technology and service, Goat Group aims to be the most trusted source for discovering, buying, and selling authentic sneakers.

Lilt

Above: Lilt

Image Credit: Lilt

Lift has a platform that boosts the work of professional translators and increases the domain expertise for specific projects with its hybrid human-machine training pipeline. The San Francisco startup’s software enables users to review each line of text in one language and translate it into another, offering entire lines of translation suggestions.

Lilt aims to delivery high-quality translations that enable companies to bring their products to more countries faster than ever. Its AI works alongside human translators to boost quality, and it uses human domain experts to better help customers.

This approach takes a human-in-the-loop path to the machine translation platform, combining neural networks with the expertise of human translators. The system is built to work like bilingual predictive text, suggesting translations and constantly learning and adapting based on what the translators accept or reject.
The company said it drastically lowers the cost and timespan of effective translation services so more businesses can expand into new markets and operate internationally. This opens new doors to the global knowledge economy and drives Lilt’s vision of making information online accessible to everyone, beyond a lingua franca.

Babblelabs

Above: Babblelabs

Image Credit: Babblelabs

Babblelabs is upgrading digitized speech with AI and may soon emerge as a big player in solving the cocktail party problem of background noise. The Campbell, California startup’s Clear Cloud API, trained on Nvidia GPUs, offers automatic speech enhancement and noise reduction for sound processing.

The company enhances speech understanding for both people and machines on platforms like phones, automobiles, appliances, and voice monitors. It also has applications in industries ranging from home internet of things (IoT) devices to customer service, education, and law enforcement.

It provides “industry-best speech quality,” intelligibility, and personalization through a unique combination of speech science, deep learning algorithms, and a vast training corpus. The resulting audio, video, speech clarity, and speech recognition in noisy environments means better device performance, higher quality service, improved safety, and overall cost reduction.

Babblelabs is building a unique technology and business around speech, AI, and optimized computing implementations. It builds unique neural network structures that are trained and tuned by speech experts to extract astonishingly clean speech from noise and implemented on GPUs, CPUs, DSPs, neural network accelerators, and micro-controllers.

This means it can fit new speech enhancement, speech recognition, and speech analytics into a wide range of common devices for greater convenience, accuracy, and privacy. Its target applications are mobile telephony, critical business communications, automotive, and speech-powered IoT.

Mapillary

Above: Mapillary

Image Credit: Mapillary

Mapillary is developing detailed maps by integrating computer vision technology with community collaboration. The Malmö, Sweden-based startup collects street-level images from cameras to build visualizations of the world in an effort to improve maps, helping cities plan their development and contributing to the development of the automotive industry.

The startup uses collaboration and computer vision to address the rapidly growing need for detailed and accurate maps. The combination of speedy advances in autonomous driving and urban development is driving the need for a scalable way of updating maps, which explains why the global digital map market is set to surge to more than $8 billion by 2025.

Unlike other map data providers, Mapillary is independent, collaborative, and on-demand. All 465 million images on the Mapillary platform have been uploaded by people, organizations, and even countries.

The imagery and data are analyzed automatically and at scale using Mapillary’s computer vision technology. Because of the company’s huge and diverse set of street-level images from 190 countries, taken with cameras ranging from smartphones to professional rigs, it has managed to build some of the computer vision technology for street scenes.

Clients include mapping companies like Here and Mapbox, automotive players like Toyota and AID (Audi’s subsidiary focusing on autonomous vehicles), and local authorities ranging from metropolitan cities like Los Angeles to small towns like Clovis in New Mexico.

Mapillary has raised some $24.5 million from Atomico, Sequoia Capital, BMW i Ventures, and Samsung Catalyst Fund. CEO Jan Erik Solem cofounded the company in 2013.

Prenav

Above: Prenav

Image Credit: Prenav

Prenav is a drone developer focused on infrastructure inspections. The Redwood City, California startup uses its home-brewed lidar and guidance system to put drones in close proximity to industrial inspection points and allow them to navigate indoors without access to GPS. Prenav’s inspections include U.S. cell phone towers and energy infrastructure.

The company created highly automated drones and deep learning algorithms for the inspection of critical infrastructure (bridges, dams, power plants, and electrical towers). That’s significant because infrastructure in the U.S. is aging, and the American Society of Civil Engineers estimates a $2 trillion funding gap for infrastructure over the 10 years from 2016 to 2025.

That means ongoing inspections and maintenance are becoming even more important, as structures are being extended well beyond their intended lifespan. Currently, the inspection of large concrete and steel structures is primarily carried out by workers on ropes, scaffolding, or bucket trucks, and the overall process is slow, expensive, dangerous, and inaccurate. As a result, critical damage and defects are sometimes missed, leading to catastrophic consequences such as the Ponte Morandi bridge collapse in Italy, the Oroville Dam crisis, and 17 wildfires in 2017 that have been attributed to PG&E equipment by Cal Fire.

Prenav’s drones fly close to structures (i.e. in GPS-denied environments like right up close, underneath, or indoors). The system leverages a unique lidar-enabled base station on the ground for GPS-denied navigation and then takes hundreds of high-resolution photos that are stitched together to build up a detailed 3D reconstruction or “digital twin” of the structure that’s being inspected. Custom deep learning algorithms then look for common damage and defects, such as cracks in concrete, fractures in steel, missing rivets, rusted cotter keys, flashed over insulators, and more.

Prenav can digitize large concrete and steel structures at 0.2 millimeter surface resolution. The system consists of a precision-guided drone for GPS-denied environments and Prenav.xyz, a web-based platform for visualizing the high-resolution digital twins and determining damage.

The company raised a $6.5 million seed round in 2016 and has recently emerged from R&D, with customers across transportation, energy, telecommunications, and construction.

Vyasa Analytics

Above: Vyasa Analytics

Image Credit: Vyasa Analytics

Vyasa’s deep learning platform enables customers to run queries in the areas of life sciences, health care, marketing, legal, and business intelligence. The Boston-based startup unleashes GPUs to help customers tap into data silos to extract insights.

The company provides a highly scalable deep learning platform, dubbed Cortex, for collaborative data sciences in project teams. With its library of analytics modules, Cortex enables novel deep learning-driven analytics of image, text, and quantitative and small compound data sources for use cases in the life sciences and health care verticals. That includes crystal morphology for formulation, high content cell assay screening, competitive intelligence, de novo compound design, and EHR analytics.

Kinetica

Above: Kinetica provides real-time data analytics for businesses.

Image Credit: Kinetica

Kinetica offers a platform that combines streaming and historical data with location intelligence and machine learning-powered analytics. Companies in the automotive, energy, telecommunications, retail, and financial services industries use the platform’s GPU-accelerated computing power to build custom analytical applications that deliver real-time results.

Kinetica’s analytics run constantly and complete updates in real time, directing other systems based on the results. The active analytics let customers build custom applications for business decisions — informed by the context of historical information, machine learning, and AI for predictive analysis.

The Kinetica Active Analytics Platform empowers enterprises to continuously and automatically combine, analyze, and immediately act on billions of live, streaming, and historical data events. With Kinetica, a business can build smart, analytical applications to enable digital business capabilities such as automated replenishment in retail, adaptive coverage for telcos, continuous risk assessment in finance, pattern recognition for AV enablement, and predictive analytics in pharmaceutical research.

The Kinetica Platform includes a distributed, in-memory, Nvidia GPU-accelerated database that utilizes a powerful combination of CPUs and GPUs to analyze massive, complex datasets with millisecond response times.

Kinetica has active analytics, dramatically simplifying the architecture to deliver smart analytical applications at massive scale, uniting historical analytics, streaming analytics, graph analytics, location intelligence, and machine learning-powered analytics.