NVAIL Partners Showcase Trailblazing Deep Learning Research at ICLR

The International Conference on Learning Representations isn’t a name that rolls off the tongue. But for researchers looking to stay on the cutting edge of deep learning, it’s the place to be.

Better known as ICLR, this year’s conference will bring together experts from the world’s top AI research labs to Vancouver from April 30-May 3. Three of our NVIDIA AI Labs (NVAIL) partners — the Swiss AI Lab (IDSIA), New York University and the University of Tokyo — are among those sharing their work.

IDSIA researchers aim to give robots the same kind of understanding of the physical world that comes naturally to people. A University of Tokyo team will discuss its innovative method for improved sound recognition. And researchers from NYU and the University of the Basque Country will explain how they’re improving machines’ ability to translate languages.

Our NVAIL program helps us keep these and other AI pioneers ahead of the curve with support for students, assistance from our researchers and engineers, and access to the industry’s most advanced GPU computing power.

What Goes Up Must Come Down

Humans innately understand the physical world. We can navigate rooms we’ve never visited. If a shoe drops, we know it’ll hit the floor. And we’re well aware we can’t walk through walls. Even infants possess some basic physical understanding.

Machines don’t have it so easy. Today, training a deep learning model to understand things like “what goes up, must come down” requires lots of data and human effort to label it, said Sjoerd van Steenkiste, a Ph.D. student at IDSIA.

He and a team of researchers from IDSIA and the University of California, Berkeley, are working to streamline that process by eliminating the need for massive data and human interaction.

In a paper for ICLR, the researchers describe how they trained a neural network without human input, a process known as unsupervised learning. Using our DGX-1 AI supercomputer, they trained a deep learning model to distinguish individual objects in a scene and predict the consequences of actions.

Eventually, this research could make it easier to train robots and other machines to interact with their environments, van Steenkiste said.

Sound Mix

Some things are just better mixed together. Peanut butter paired with chocolate is heavenly. Metals are stronger and harder when they’re combined. And planting two crops together can yield bigger harvests.

Yuji Tokozume is applying the same idea to deep learning. The doctoral student and two other University of Tokyo researchers are set on improving sound recognition by using what they call between-class sounds — two sounds mixed together — to train a deep learning model. The model, trained on our Tesla P100 GPU accelerators, identifies the two sounds and determines the ratio of the one sound to another.

In their ICLR paper, the researchers report that between-class learning not only delivered higher accuracy than existing techniques but also surpassed human performance on environmental recordings in a standard dataset known as ESC-50. The team has applied the same approach to improve AI image recognition performance.

Learn more by viewing a talk on between-class learning for sound recognition at our recent GPU Technology Conference in Silicon Valley.

Lost in Translation

For all AI has achieved in automatic language translation, it doesn’t do much for less common tongues like Basque, Oromo and Quechua. That’s because training a deep learning model typically requires large datasets — in this case, vast amounts of text that’s been manually translated into other languages.

Ample data for widely spoken languages like Chinese, English and Spanish makes it possible to directly translate Chinese to English or Spanish to Chinese. Researchers at NYU and the University of the Basque Country aim to bring that capability to languages with smaller numbers of speakers.

Currently languages like Basque — spoken by an estimated 700,000 people, mostly in a region that straddles Spain and France — must first be translated into English (or another major language) before they can be converted to anything else, according to Mikel Artetxe, a doctoral student at the University of the Basque Country.

The same holds true for languages such as Oromo, which is spoken by more than 30 million people in the Horn of Africa, or Quechua, which is spoken by as many as 11 million people in South America.

The research team used our TITAN Xp GPUs to train a neural network to perform these translations without any manually translated training data, relying on independent text of both languages instead. In their ICLR paper, researchers said that accuracy improved when they added a small amount of parallel data, although it was still far below that of a human translation.

“Our goal is to be able to translate more languages with better results,” said Artexe.

The post NVAIL Partners Showcase Trailblazing Deep Learning Research at ICLR appeared first on The Official NVIDIA Blog.

NVIDIA Research Pushes Deep Learning Forward at ICLR

NVIDIA researchers grabbed headlines last fall for generating photos of believable yet imaginary celebrity faces with deep learning. They’ll discuss how they did it on stage next week at the International Conference on Learning Representations, better known as ICLR.

That research team is one of five from NVIDIA Research sharing their work to advance deep learning at ICLR, April 30-May 3 in Vancouver. Our 200-person strong NVIDIA Research team, which works from 11 worldwide locations, focuses on pushing the boundaries of technology in machine learning, computer vision, self-driving cars, robotics, graphics and other areas.

Also at ICLR: The NVIDIA Deep Learning Institute will offer free online training — and the chance to win an NVIDIA TITAN V. (More information below.) And, if you missed our GPU Technology Conference, you can see our latest innovations at our booth.

ICLR, in its sixth year, is focused on the latest deep learning techniques. NVIDIA is a sponsor of the conference.

More Than a Pretty Face

In the face-generating research, a team at our Finland lab developed a method of training generative adversarial networks (GANs) that produced better results than existing techniques. The researchers demonstrated their success by applying it to the difficult problem of generating realistic-looking human faces.

“Human looks have been somewhat sacred. It’s extremely difficult to create believable-looking digital characters in, say, movies without using real-life actors as reference,” said Tero Karras, lead author on the ICLR paper. “With deep learning and this paper, we’re getting closer.”

The neural network generates new human faces by mixing characteristics like gender, hairstyle, and face shape in different ways. The video below shows the result of varying these characteristics at random, demonstrating an endless number of possible combinations.

The research paves the way for game developers to more quickly and easily create digital people who look like the real thing, Karras said. He’s also heard from a team that’s looking to apply the research to help people with prosopagnosia, or face blindness, a neural disorder in which sufferers can’t recognize faces.

The researchers will discuss the paper, “Progressive Growing of GANs for Improved Quality, Stability, and Variation,” on Monday morning at ICLR, explaining how they achieved such good results and the reasoning behind their techniques.

Poster Sessions

Check out our poster sessions at ICLR:

  • Deep Gradient Compression: Reducing the Communication Bandwidth for Distributed Training” – Training a neural network across many machines is an efficient way to train deep and larger models, but it requires an expensive high-bandwidth communications network. This research makes it possible train models faster across more GPUs, but on an inexpensive commodity Ethernet network.
  • Sparse Persistent RNNs: Squeezing Large Recurrent Networks On-Chip” – Engineers combined a method for more efficiently running recurrent neural networks on GPUs with a technique known as model pruning, which reduces the complexity of a neural network. The result dramatically speeds up models and makes it possible to deploy far larger neural networks on each GPU.
  • Efficient Sparse-Winograd Convolutional Neural Networks” – Convolutional neural networks are computationally intensive, which makes it difficult to deploy them on mobile devices. Researchers made changes to what’s known as the Winograd-based convolution algorithm — a method used to reduce the number of computations needed to process CNNs — so that it would work with network pruning, a way to reduce parameters and increase speed in a neural network. In addition to shrinking computational workloads, combining the two methods allowed researchers to prune networks more aggressively.
  • Mixed Precision Training” – Increasing the size of a neural network typically improves accuracy but also increases the memory and compute requirements for training the model. A team from NVIDIA and Baidu developed a new training method called mixed-precision training, which halves memory requirements and shortens training or inference time — without losing accuracy or having to change parameters.

Get Free Training, Win a TITAN V

The NVIDIA Deep Learning Institute (DLI) will offer free online training exclusively to ICLR attendees — and the chance to win an NVIDIA TITAN V. Stop by NVIDIA booth 100 to pick up a token card for free access. The first 200 attendees to take an online course will receive $100 in online training credits. See contest rules. Plus, visit our booth on Tuesday and Wednesday between 4-5 p.m. to meet our DLI University Ambassadors.

Also at our booth: get hands on with our some of our latest technology, talk with deep learning experts or meet our hiring team.

The post NVIDIA Research Pushes Deep Learning Forward at ICLR appeared first on The Official NVIDIA Blog.

Doctors Are Superheroes Today, Superhumans Tomorrow

AI is the most important technology of our time, while early detection is the most important medical challenge of our time. Incredible breakthroughs in AI are making it possible for doctors to see disease earlier and better understand it.

That’s why NVIDIA founder and CEO Jensen Huang is speaking at the World Medical Innovation Forum (WMIF) this week in Boston. More than 1,000 worldwide leaders in industry and academia will share recent innovations at the intersection of AI and healthcare — this is the focus of Huang’s fireside chat.

NVIDIA’s Medical Record

NVIDIA’s work in virtual reality gaming, AI and self-driving cars constantly grabs headlines. Hidden in our chart is the great work we do with partners in healthcare.

The earliest applications of CUDA, our GPU computing platform, were in medical imaging and life science. Invented 15 years earlier, iterative reconstruction was a new algorithm in computed tomography that promised X-ray dose reduction but was too computationally expensive. NVIDIA GPUs made it possible for GE Healthcare’s family of Revolution CT to reduce radiation by 82 percent.

Ultrasound has also been revolutionized by GPU computing; GE Healthcare’s Vivid E95 can perform real-time 4D imaging to see blood flow through the heart.

NVIDIA GPU computing enabled life science researchers at Klaus Schulten’s computational biophysics lab at the University of Illinois at Urbana-Champaign to simulate molecular dynamics at a scale 1,000x larger than otherwise possible. This allowed them to create never seen before views of an HIV capsid and the first ever simulation of an entire life form, that of the satellite tobacco mosaic virus.

ThermoFisher Scientific engineers accelerated their gene sequencing algorithm by 250x with NVIDIA GPUs, and created the Ion S5 Next Generation Sequencing system — a breakthrough in both cost and time savings to analyze a targeted gene sequence.

At the University of Stockholm, NVIDIA GPUs enable researcher Erik Lindahl’s RELION imaging application to process and reconstruct 3D images of molecular structures. RELION is the imaging software of the Nobel Prize-winning cryogenic electron microscope. Cryo-EM lets researchers freeze molecules in mid-motion and see biological processes at the atomic level for the first time. The journal Nature dubbed Cryo-EM its scientific “Method of the Year.”

Healthcare has been an intense focus at NVIDIA for over a decade.

AI Give Doctors Superhuman Powers

Deep learning burst onto the computing scene six years ago when Alex Krizhevsky won ImageNet, the international computer image recognition contest. Krizhevsky used NVIDIA GPUs to train his eight-layer deep neural network, called AlexNet.

Recognizing the importance of this approach of software development, NVIDIA went all-in on deep learning, believing it would lead to advances in AI and help solve problems never before possible. It was a good decision. Today, deep learning software has achieved superhuman pattern recognition capabilities — in vision, sound, speech and many other forms of perception.

This year, we announced the new Volta GPU — the first processor that is equally adept at computer graphics, scientific computing and deep learning. With Volta’s Tensor Core architecture, AlexNet can be trained 500x faster than just six years ago. NVIDIA is advancing AI computing at lightning speeds.

Medical imaging researchers have discovered the power of deep learning. Half of the papers presented at last year’s MICCAI, the leading medical imaging conference, applied deep learning. We’re working with over 300 healthcare startups tackling challenges now possible with deep learning. Together they’ve raised over $1.5 billion. Arterys, Butterfly Networks, RADLogics, Viz.Ai and Zebra are doing exciting work in medical imaging, with many AI recognition models that are now FDA approved.

The progress is amazing, and this is just the first stage.

The Future

AI is truly great with promise to augment future superhuman doctors. The pace of progress is incredible.

How do we get this technology into the hands of doctors? And how do we integrate this technology into current infrastructure, which was created with great care to protect patient data?

We recently announced a new GPU computing platform, called Clara, that accelerates imaging of different modalities, neural network architectures of all types and any approach of visualization — from rasterization, to volumetric, to ray tracing. Clara is built for the datacenter, extending our offering from embedded in medical devices, workstations, on-prem datacenters, or any and every cloud. Clara can be used to run the latest breakthroughs in medical imaging, virtually, and upgrade the world’s 3 million medical instruments instantly.

It’s one architecture — with the same software that can run everywhere.

Clara medical imaging supercomputer

We partner with leaders in healthcare. In medical imaging, we have great partnerships with GE, Siemens, Philips and Nuance. We recently announced a new partnership with Canon Medical Systems to put AI supercomputers in hospitals. We partner with leading research hospitals — including Massachusetts General Hospital, Brigham and Women’s Hospital, and Mayo Clinic — the National Institutes of Health and startups like Paige.AI and PathAI, which are working on AI-powered computational pathology.

All of us have one purpose — to empower future doctors with superhuman capabilities so that better healthcare can be provided.

The post Doctors Are Superheroes Today, Superhumans Tomorrow appeared first on The Official NVIDIA Blog.

NVIDIA, SoftBank Incubator DEEPCORE Team Up to Fuel AI Startups in Japan

NVIDIA and DEEPCORE, a Tokyo-based startup incubator owned by SoftBank, are working together to support AI startups and promote university research programs across Japan.

Launched earlier this year with a mission to cultivate entrepreneurs who aspire to change the world with technology, DEEPCORE will use NVIDIA’s AI computing platform to build out the technology infrastructure of its incubation program.

Program members will have access to NVIDIA DGX systems and the NVIDIA GPU Cloud, which gives researchers and data scientists easy access to a comprehensive catalog of GPU-optimized software tools for deep learning and high performance computing. DEEPCORE is developing its GPU-accelerated AI computing platform in its open, collaboration-focused R&D space, called KERNEL, located near the University of Tokyo.

NVIDIA will offer AI training to DEEPCORE customers and incubator members via its Deep Learning Institute, which provides hands-on training for developers, data scientists and researchers.

NVIDIA will also provide technical and business advice to members regarding its GPU-accelerated products and services.

DEEPCORE and NVIDIA additionally plan to explore business opportunities for members as part of NVIDIA’s Inception program, a virtual startup accelerator working with more than 2,800 companies around the world.

DEEPCORE works in collaboration with the University of Tokyo and other top schools to unleash the potential in technology for entrepreneurs, researchers, engineers and others in Japan and around the globe.

 

The post NVIDIA, SoftBank Incubator DEEPCORE Team Up to Fuel AI Startups in Japan appeared first on The Official NVIDIA Blog.

iNaturalist: An AI-Powered App to Crow About on Earth Day

This April 22 put a little AI in your Earth Day.

Whether you’re cleaning up a beach, planting a tree or starting a garden, iNaturalist makes it easy to get in touch with nature.

iNaturalist is a crowd-sourced species identification app powered by AI. For the casual nature observer, the app allows people to snap photos of such easy targets as backyard plants and bugs and upload images for its AI to provide a match or for members of the community to identify.

The app is also a social network for nature enthusiasts to record information on species, meet others with similar interests and learn. It’s available on Android and iOS , where it’s already been downloaded nearly a million times.

It began as a website, iNaturalist.org, founded in 2008 by students at the University of California, Berkeley. Now it’s a joint program of the California Academy of Sciences and the National Geographic Society.

Previously it took 18 days on average for species to be identified by the website’s community. But that all changed after iNaturalist worked with researchers from Caltech and Cornell to build a computer vision AI into the app.

Now species can be identified in a matter of milliseconds, and much more accurately, with the use of AI, said Scott Loarie, co-director of iNaturalist.

“Our goal is to get millions of people outside exploring and connecting to nature and engage them to become lifelong stewards of the natural world,” Loarie said.

People mostly put up observations of plants to iNaturalist, but posts of birds, insects and other organisms can be found as well.

The app harnesses NVIDIA GPUs and the CUDA deep neural network library along with the TensorFlow deep learning framework, allowing training of the neural networks on a database of images that have been labeled by the site’s community of experts.

Today iNaturalist boasts 8.6 million observations uploaded and more than 155,000 species observed.

The post iNaturalist: An AI-Powered App to Crow About on Earth Day appeared first on The Official NVIDIA Blog.

It’s Training Cats and Dogs: NVIDIA Research Uses AI to Turn Cats into Dogs, Lions and Tigers, Too

It turns out a leopard can change its spots.

Thanks to NVIDIA researchers’ new GPU-accelerated deep learning technique, a leopard — or at least a picture of it — can simultaneously turn into a house cat, a tiger or even a dog. It works for video, too.

The ability to turn one image or video into many could help game developers and filmmakers move faster, spend less and create richer experiences. It also could boost autonomous vehicles’ ability to handle a wider variety of road conditions by more quickly and easily generating diverse training data.

One to Many

With this discovery, the researchers one-up their earlier work on image translation, presented in December at the Conference and Workshop on Neural Information Processing Systems, better known as NIPS. The method described in the NIPS paper worked on a one-by-one basis, mapping one image or video to another.

The new technology, disclosed in a paper published today, is “multimodal” — it simultaneously converts one image to many.

Multimodal image translation is just the latest example of groundbreaking work from our 200-person strong NVIDIA Research team. Spread across 11 worldwide locations, our researchers are pushing the boundaries of technology in machine learning, computer vision, self-driving cars, robotics, graphics, computer architecture, programming systems and other areas.

Sunshine on a Cloudy Day

Like the NIPS research, multimodal image translation relies on two deep learning techniques — unsupervised learning and generative adversarial networks (GANs) — to give machines more “imaginative” capability, such as imaging how a sunny street would look during a rainstorm or in wintertime.

Now, instead of translating a summer driving video into just one instance of winter, the researchers can create a diverse set of winter driving videos in which the amount of snow varies. The technology works the same way for different times of day and other weather conditions, providing sunshine on a cloudy day or turning darkness to dawn, afternoon or twilight. This technique can be extremely valuable in training deep neural networks that can be used in self-driving cars.

In the world of gaming, multimodal image translation could give studios a faster and easier way to create new characters or new worlds. Artists could set aside more tedious tasks to develop richer and more complex stories.

The Multimodal Unsupervised Image-to-image Translation framework, dubbed MUNIT, works by separating image content from style. In a picture of a cat, for example, the pose of the cat is the content and the breed is the style. The pose is fixed. If you’re converting a picture of a housecat to a leopard or a dog, the position of the animals must remain identical. What varies is the breed or species — domestic shorthair, leopard or collie, for example.

No Data? No Problem

This research is built on a deep learning method that’s good at generating visual data. A GAN uses two competing neural networks — one to generate images and one to assess whether the generated images are real or fake. GANs are particularly useful when there’s a shortage of data.

Typically, image translation requires datasets of corresponding images — pictures of collies, labrador retrievers or tigers positioned in the exact same way as an original cat image. That sort of data is difficult, if not impossible, to find. The advantage of MUNIT is that it works without it.

MUNIT could also be handy for generating training data for autonomous cars by eliminating the need to capture the same footage recorded from the same vantage point, with the same perspective and with all the oncoming traffic and other details in exact same location.

In addition, GANs eliminate the need for people to label the content of each image or video, a task that takes extensive time and manpower.

“My goal is to enable machines to have human-like imaginative capabilities,” said Ming-Yu Liu, one of the authors of the paper. “A person can imagine what a scene looks like in wintertime, whether the trees are bare of leaves or covered with snow. I hope to develop artificial intelligence that can do this.”

The post It’s Training Cats and Dogs: NVIDIA Research Uses AI to Turn Cats into Dogs, Lions and Tigers, Too appeared first on The Official NVIDIA Blog.

Hail Yes: How Deep Learning Could Improve Forecasts for Damaging Storms

Imagine tens of thousands of golf balls falling from the sky at more than 100 mph, and you’ll get an idea of the damage hail can wreak.

In just a few minutes, it can batter and ruin crops, dent cars and smash windshields, and even bash holes in houses and buildings — resulting in billions of dollars in losses.

“Because hail is so damaging, we want to be able to forecast it more reliably so people can take shelter and protect their property,” said David Gagne II, a postdoctoral fellow at the National Center for Atmospheric Research, in a talk at the GPU Technology Conference this week.

Gagne and other scientists at NCAR are using GPU-accelerated deep learning to more accurately forecast the likelihood of hail, where it will fall and how large it will be.

Current Hail Forecasts Fall Short

Hail happens when upward air currents from thunderstorms are strong enough to carry water droplets well above freezing level. These frozen droplets become hailstones, which grow as additional water freezes to them. When hailstones are too heavy for the updrafts, they fall to the ground.

Meteorologists and other scientists have multiple ways of predicting storms, but all of these have shortcomings that can result in missed storms and false alarms, Gagne said. Scientists have also experimented with machine learning-based forecasts.

“Machine learning can produce reliable severe weather forecasts but it struggles with learning spatial patterns,” Gagne said. These patterns show what areas will be affected by rain or hail.

AI Prediction Potential

By contrast, it’s relatively easy to integrate spatial patterns, time and physical understanding of conditions into deep learning models, according a paper by Gagne and other scientists published in the American Meteorological Society journal.

AI also has the potential to reveal new knowledge in data such as Doppler radar maps, the multi-colored maps you see in weather forecasts on TV.

“I wonder if deep learning can look at these images and see what meteorologists see, or if it sees something different,” Gagne said.

Predicting Hail That Dings Your Car

He and the team use NVIDIA Tesla GPUs and the cuDNN-accelerated TensorFlow deep learning framework to train their models to predict hail more than 25 millimeters (about an inch) in diameter, or about the size of a quarter.

“That’s the size where it will ding your car or mess up your roof,” Gagne said.

In experiments so far, their models generally produced fewer false alarms and higher accuracy than other methods. Better hail predictions would give people the chance to move to protected areas, park their cars beyond the reach of storms and allow airports to reroute planes or cancel flights, Gagne said.

Gagne and other scientists are also experimenting with AI to predict the type of expected precipitation, severe winds and storm lifetimes.

 

The post Hail Yes: How Deep Learning Could Improve Forecasts for Damaging Storms appeared first on The Official NVIDIA Blog.

Groundbreaking Deep Learning Research Takes the Stage at GTC

NVIDIA researchers aren’t magicians, but you might think so after seeing the work featured in today’s keynote address from NVIDIA founder and CEO Jensen Huang at the GPU Technology Conference, in San Jose.

Huang spotlighted two deep learning discoveries that hold the potential to upend traditional computer graphics. Both could help game developers create richer experiences in less time and at lower cost. And one could accelerate autonomous vehicle development by easily creating data to train cars for a wider variety of road conditions, landscapes and locations.

The pair of research projects are the latest examples of how we’re combining our expertise in deep learning with our long history in computer graphics to advance industries. Our 200-person strong NVIDIA Research team — spread across 11 worldwide locations — is focused on pushing the boundaries of technology in machine learning, computer vision, self-driving cars, robotics, graphics, computer architecture, programming systems, and other areas.

“The productivity of this organization is absolutely incredible,” Jensen said. “They’re doing fundamental and basic research across the entire stack of computing.”

De-noising the image on the left was done using the traditional method, training a neural network on corresponding clean and noisy images. The NVIDIA researchers' image on the right was created by training the network on noisy images only.
The two images here are clean versions of the same noisy picture. De-noising the image on the left was done by training a neural network on corresponding clean and noisy images. Researchers de-noised the picture on the right using a model trained soley on noisy images.

Cleaning up Noisy Images

You may not know what a noisy image is, but you’ve probably taken one. You aim the camera at a dimly lit scene, and your picture turns out grainy, mottled with odd splotches of color, or white spots known as fireflies.

Removing noise from images is difficult because the process itself can add artefacts or blurriness. Deep learning experiments have offered solutions, but have a major shortcoming: They require matched pairs of clean and noisy images to train the neural network.

Ordinary AI denoising requires matched pairs of clean and dirty images. But it's often impossible to get clean images for MRIs and some other medical images. With Noise2Noise, no clean images are necessary.
Ordinary AI denoising requires matched pairs of clean and dirty images. But it’s often impossible to get clean images for MRIs and some other medical images. With Noise2Noise, no clean images are necessary.

That works as long as you have good pictures, but it can be hard, or even impossible, to get them. NVIDIA researchers in Finland and Sweden have created a solution they call Noise2Noise to get around this issue.

Garbage in, Garbage out? Not Anymore

Producing clean images is a common problem for medical imaging tests like MRIs and for astronomical photos of distant stars or planets —  situations in which there’s too little time and light to capture a clean image.

Time also poses a problem in computer graphics. Just the task of generating clean image data to train a denoiser can take days or weeks.

Noise2Noise seems impossible when you first hear about it. Instead of training the network on matched pairs of clean and noisy images, it trains the network on matched pairs of noisy images — and only noisy images. Yet Noise2Noise produces results equal to or nearly equal to what a network trained the old-fashioned way can achieve.

“What we’ve discovered is by setting up a network correctly, you can ask it do something impossible,” said David Luebke, our vice president of research. “It’s a really surprising result until you understand the whole thing.”

Not Child’s Play

The second project Huang featured represents a whole new way of building virtual worlds. It uses deep learning to take much of the effort out of cumbersome and costly tasks of 3D modeling for games and capturing training data for self-driving cars.

The technique, called semantic manipulation, can be compared to Lego bricks, which kids can put together to build anything from jet planes to dragons.

In semantic manipulation, users start with a label map — what amounts to a blueprint with labels for each pixel in a scene. Switching out the labels on the map changes the image. It’s also possible to edit the style of objects, like choosing a different kind of car, tree or road.

The NVIDIA researchers' AI-powered image synthesis technique makes it possible to change the look of a street simply by changing the semantic label.
The NVIDIA researchers’ deep learning-powered image synthesis technique makes it possible to change the look of a street simply by changing the semantic label.

Tough Game

The research team’s method relies on generative adversarial networks (GANs), a deep learning technique often used to create training data when it’s scarce.

Although GANs typically struggle to generate photorealistic, high-resolution images, NVIDIA researchers were able to alter GAN architecture in a way that made it possible.

Today, creating virtual environments for computer games requires thousands of hours of artists’ time to create and change models and can cost as much as $100 million per game. Rendering turns those models into the games you see on the screen.

Reducing the amount of labor involved would let game artists and studios create more complex games with more characters and story lines.

San Francisco to Barcelona: No Flight Required  

Obtaining data to train self-driving cars is equally cumbersome. It’s typically done by putting a fleet of cars equipped with sensors and cameras on the road. The data captured by the cars must then be labeled manually, and that’s used to train autonomous vehicles.

The team’s method could make it possible to take data from, say, San Francisco and apply it to another hilly city like Barcelona. Or turn a cobblestone street into a paved one, or convert a tree-lined street into one lined with parked cars.

That could make it possible to more effectively to train cars to handle many different situations. And could lead to a graphics rendering engine that’s been trained on real-world data and rendered with generative models.

“I’m so proud of our NVIDIA research team,” Jensen said. “We’re growing. Reach out to us. We’d love to work with you.”

For more information about how our researchers are revolutionizing graphics, see the papers (listed below) or read our related articles, “NVIDIA Research Brings AI to Graphics” and “NVIDIA Researchers Showcase Major Advances in Deep Learning at NIPS.”

 

The post Groundbreaking Deep Learning Research Takes the Stage at GTC appeared first on The Official NVIDIA Blog.

Why Deep Learning May Prove to Be the Bee’s Knees

In Bee Movie, an animated feature from 2007, a friendship between a bee (voiced by Jerry Seinfeld) and a young woman (Renee Zellweger) leads to the world’s bee population reclaiming the honey it produces.

A decade later, a young woman who is real and with a self-described “penchant for cute, round things” — working with NVIDIA engineers and GPU-powered deep learning — may help to minimize the impact of a destructive parasite and lead to domesticated bees being returned to the almond-shape hive design that serves them so well in the wild.

Jade Greenberg, a 17-year-old junior at Pascack Hills High School in Bergen County, N.J., zeroed in on honey bees and the causes of colony collapse as the subject of a research project for her molecular genetics class.

Eventually, Greenberg focused on the threat posed by Varroa mites, a parasite thought to be one of the most frequent causes of collapses of domestic hives. Her research has led her to postulate that the long-accepted design of the Langstroth hive — the cabinet-like standard since its introduction in the 1850s — is a big reason Varroa mites have become such a big problem.

The classic Langstroth hive design.

NVIDIA, GPUs and deep learning came into the picture when Jade’s father, a solutions engineer at Kinetica, teamed with Jacci Cenci, an NVIDIA engineer, and they started applying their companies’ respective technologies to the problem. Linking sensors and cameras to a convolutional neural network, Cenci’s team began collecting data on hive conditions such as weight, humidity, temperature and population.

Ramping Up Detection

A variety of deep learning and machine learning technologies — including NVIDIA’s Jetson TX2 development kit, an NVIDIA DGX Station, TensorRT, a high-performance deep learning inference optimizer and the Microsoft Cognitive Toolkit deep learning framework — combine to rapidly detect and warn against the presence of Varroa mites.

The stack of NVIDIA hardware and software is able to optimize, validate and deploy trained neural networks for inferencing in the field, thereby alerting beekeepers of the potential for infestation sooner.

“If the weight decreases, the hive could be sick and bees are leaving. If the hive is heavy, it could mean lots of swarming, or high humidity might be present, which increases the odds of mite infestation,” said Cenci.

Armed with this extra information, which is converted into useful charts and graphs using Kinetica’s GPU-accelerated insight engine, Greenberg moved steadily from simply studying the problem to crafting a solution.

“We have better ways of collecting data, and we have better ways of observing bees in more detailed ways,” she said. “I’m surprised that this industry hasn’t moved past something designed in the 19th century. It’s time for a change.”

Fighting Mites with Hive Design

Greenberg, who explains her work in a video below, has been using the data collected with the NVIDIA technology to guide her efforts to design a better hive. She’s learned that the different sizes and shapes of the entrances, the contrast with the natural almond shape of wild hives, and the fact that the queen is separated from the rest of the colony are potentially fatal flaws of the Langstroth hive design.

She also suggested that the larger spaces that bees occupy allow other mite-infected insects, such as moths, to enter the hive and become trapped, leading to further infestation.

In other words, AI is enabling Greenberg to pull back the curtain on the Langstroth hive’s failings, which may have been underestimated before now.

“It tells us in what ways the Langstroth hive is failing us when it comes to Varroa mite infestation,” she said.

It also is helping Greenberg refine her design, which, to be viable in the commercial beekeeping arena, must improve hive health while preserving the commercial capabilities of the Langstroth hive.

Greenberg and her bee hive design were recently awarded first alternate in the engineering category at the Nokia Bell Labs North Jersey Regional Science Fair. She’s also a finalist in the Intel International Science and Engineering Fair in May.

Her work and the technology backing her up will additionally be the subject of a session at the upcoming GPU Technology Conference in San Jose. Kinetica dashboards presenting the info will be rendered on an NVIDIA DGX Station AI supercomputer.

The post Why Deep Learning May Prove to Be the Bee’s Knees appeared first on The Official NVIDIA Blog.

Kyoto University Chooses Intel Machine and Deep Learning Tech to Tackle Drug Discovery, Medicine and Healthcare Challenges

Kyoto University Graduate School of Medicine*, one of Asia’s leading research-oriented institutions, has recently chosen Intel® Xeon® Scalable processors to power its clinical genome analysis cluster and its molecular simulation cluster. These clusters will aid in Kyoto’s search for new drug discoveries and help reduce research and development costs.

The Intel Xeon Scalable platform offers potent performance for all types of artificial intelligence (AI). Intel’s optimizations for popular deep learning frameworks have produced up to 127 times1 performance gains for deep learning training and 198 times2 performance gains for deep learning inference for AI workloads running on Intel Xeon Scalable processors. Kyoto is one of many leading healthcare providers and research institutions that are working with Intel and using Intel artificial intelligence technology to tackle some of the biggest challenges in healthcare.

More: One Simple Truth about Artificial Intelligence in Healthcare: It’s Already Here (Navin Shenoy Editorial) | Shaping the Future of Healthcare through Artificial Intelligence (Event Video Replay) | Artificial Intelligence (Press Kit) | Advancing Data-Driven Healthcare Solutions (Press Kit)

“We are only at the beginning of solving these problems. We are continuing to push forward and work with industry-leading entities to solve even more,” said Arjun Bansal, vice president of the Artificial Intelligence Products Group and general manager of Artificial Intelligence Labs and Software at Intel Corporation. “For example, I’m happy to announce that Kyoto University has recently chosen Intel to power their clinical genome analysis cluster and their molecular simulation cluster. These clusters are to aid in their drug discovery efforts and should help reduce the R&D costs of testing different compounds and accelerate precision medicine by adopting Deep Learning techniques.”

It can take up to 15 years – and billions of dollars – to translate a drug discovery idea from initial inception to a market-ready product. Identifying the right protein to manipulate in a disease, proving the concept, optimizing the molecule for delivery to the patient, carrying out pre-clinical and clinical safety and efficacy testing are all essential, but ultimately the process takes far too long today.

Dramatic shifts are needed to meet the needs of society and a future generation of patients. Artificial intelligence presents researchers with an opportunity to do R&D differently – driving down the resources and costs to develop drugs and bringing the potential for a substantial increase in new treatments for serious diseases.

1 Performance estimates were obtained prior to implementation of recent software patches and firmware updates intended to address exploits referred to as “Spectre” and “Meltdown.”  Implementation of these updates may make these results inapplicable to your device or system.

Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors.  Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions.  Any change to any of those factors may cause the results to vary.  You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more complete information visit http://www.intel.com/benchmarks

Source: Intel measured as of June 2017 Optimization Notice: Intel’s compilers may or may not optimize to the same degree for non-Intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice.

Configurations for Inference throughput

Processor :2 socket Intel(R) Xeon(R) Platinum 8180 CPU @ 2.50GHz / 28 cores HT ON , Turbo ON Total Memory 376.46GB (12slots / 32 GB / 2666 MHz).CentOS Linux-7.3.1611-Core , SSD sda RS3WC080 HDD 744.1GB,sdb RS3WC080 HDD 1.5TB,sdc RS3WC080 HDD 5.5TB , Deep Learning Framework caffe version: f6d01efbe93f70726ea3796a4b89c612365a6341 Topology :googlenet_v1 BIOS:SE5C620.86B.00.01.0004.071220170215 MKLDNN: version: ae00102be506ed0fe2099c6557df2aa88ad57ec1 NoDataLayer. Measured: 1190 imgs/sec vs Platform: 2S Intel® Xeon® CPU E5-2699 v3 @ 2.30GHz (18 cores), HT enabled, turbo disabled, scaling governor set to “performance” via intel_pstate driver, 256GB DDR4-2133 ECC RAM. CentOS Linux release 7.3.1611 (Core), Linux kernel 3.10.0-514.el7.x86_64. OS drive: Seagate* Enterprise ST2000NX0253 2 TB 2.5″ Internal Hard Drive. Performance measured with: Environment variables: KMP_AFFINITY=’granularity=fine, compact,1,0‘, OMP_NUM_THREADS=36, CPU Freq set with cpupower frequency-set -d 2.3G -u 2.3G -g performance. Deep Learning Frameworks: Intel Caffe: (http://github.com/intel/caffe/), revision b0ef3236528a2c7d2988f249d347d5fdae831236. Inference measured with “caffe time –forward_only” command, training measured with “caffe time” command. For “ConvNet” topologies, dummy dataset was used. For other topologies, data was stored on local storage and cached in memory before training. Topology specs from https://github.com/intel/caffe/tree/master/models/intel_optimized_models (GoogLeNet, AlexNet, and ResNet-50), https://github.com/intel/caffe/tree/master/models/default_vgg_19 (VGG-19), and https://github.com/soumith/convnet-benchmarks/tree/master/caffe/imagenet_winners (ConvNet benchmarks; files were updated to use newer Caffe prototxt format but are functionally equivalent). GCC 4.8.5, MKLML version 2017.0.2.20170110. BVLC-Caffe: https://github.com/BVLC/caffe, Inference & Training measured with “caffe time” command.  For “ConvNet” topologies, dummy dataset was used. For other topologies, data was st ored on local storage and cached in memory before training  BVLC Caffe (http://github.com/BVLC/caffe), revision 91b09280f5233cafc62954c98ce8bc4c204e7475 (commit date 5/14/2017). BLAS: atlas ver. 3.10.1.

2Configuration for training throughput:

Processor :2 socket Intel(R) Xeon(R) Platinum 8180 CPU @ 2.50GHz / 28 cores HT ON , Turbo ON Total Memory 376.28GB (12slots / 32 GB / 2666 MHz).CentOS Linux-7.3.1611-Core , SSD sda RS3WC080 HDD 744.1GB,sdb RS3WC080 HDD 1.5TB,sdc RS3WC080 HDD 5.5TB , Deep Learning Framework caffe version: f6d01efbe93f70726ea3796a4b89c612365a6341 Topology :alexnet BIOS:SE5C620.86B.00.01.0009.101920170742 MKLDNN: version: ae00102be506ed0fe2099c6557df2aa88ad57ec1 NoDataLayer. Measured: 1023 imgs/sec vs Platform: 2S Intel® Xeon® CPU E5-2699 v3 @ 2.30GHz (18 cores), HT enabled, turbo disabled, scaling governor set to “performance” via intel_pstate driver, 256GB DDR4-2133 ECC RAM. CentOS Linux release 7.3.1611 (Core), Linux kernel 3.10.0-514.el7.x86_64. OS drive: Seagate* Enterprise ST2000NX0253 2 TB 2.5″ Internal Hard Drive. Performance measured with: Environment variables: KMP_AFFINITY=’granularity=fine, compact,1,0‘, OMP_NUM_THREADS=36, CPU Freq set with cpupower frequency-set -d 2.3G -u 2.3G -g performance. Deep Learning Frameworks: Intel Caffe: (http://github.com/intel/caffe/), revision b0ef3236528a2c7d2988f249d347d5fdae831236. Inference measured with “caffe time –forward_only” command, training measured with “caffe time” command. For “ConvNet” topologies, dummy dataset was used. For other topologies, data was stored on local storage and cached in memory before training. Topology specs from https://github.com/intel/caffe/tree/master/models/intel_optimized_models (GoogLeNet, AlexNet, and ResNet-50), https://github.com/intel/caffe/tree/master/models/default_vgg_19 (VGG-19), and https://github.com/soumith/convnet-benchmarks/tree/master/caffe/imagenet_winners (ConvNet benchmarks; files were updated to use newer Caffe prototxt format but are functionally equivalent). GCC 4.8.5, MKLML version 2017.0.2.20170110. BVLC-Caffe: https://github.com/BVLC/caffe, Inference & Training measured with “caffe time” command.  For “ConvNet” topologies, dummy dataset was used. For other topologies, data was st ored on local storage and cached in memory before training  BVLC Caffe (http://github.com/BVLC/caffe), revision 91b09280f5233cafc62954c98ce8bc4c204e7475 (commit date 5/14/2017). BLAS: atlas ver. 3.10.1.

The post Kyoto University Chooses Intel Machine and Deep Learning Tech to Tackle Drug Discovery, Medicine and Healthcare Challenges appeared first on Intel Newsroom.