NVIDIA CEO Jensen Huang Unveils Turing, Reinventing Computer Graphics

Ray-traced graphics offer incredible realism. Interactive graphics driven by GPUs offer speed and responsiveness. The two now come together In the greatest leap since the invention of the CUDA GPU in 2006, NVIDIA CEO Jensen Huang announced Monday.

Speaking at the SIGGRAPH professional graphics conference in Vancouver, Huang unveiled Turing, NVIDIA’s eighth-generation GPU architecture. He also introduced the first Turing-based GPUsthe NVIDIA Quadro RTX 8000, Quadro RTX 6000 and Quadro RTX 5000. And he detailed the Quadro RTX Server, a reference architecture for the $250 billion visual effects industry.

“This fundamentally changes how computer graphics will be done, it’s a step change in realism,” Huang told an audience of more than 1,200 graphics pros gathered at the sleek glass and steel Vancouver Convention Center, which sits across a waterway criss-crossed by cruise ships and seaplanes from the stunning North Shore mountains.

DellEMC, HPI, Hewlett-Packard Enterprise, Lenovo, Fujitsu, Boxx, and SuperMicro will be among the system vendors supporting the latest line of Quadro processors, he said. All three new Quadro GPUs will be available in the fourth quarter.

New Silicon, New Software

Turing — the result of more than 10,000 engineering-years of effort — features new RT Cores to accelerate ray tracing and new Tensor Cores for AI inferencing. Together for the first time, Huang explained, they make real-time ray tracing possible.

“It has to be amazing at today’s applications, but utterly awesome at tommorrow,” Huang said.

That new silicon is being supported by software from more than two dozen key ISVs. To help developers quickly take full advance of Turing’s capabilities, NVIDIA has enhanced its RTX development platform with new AI, ray-tracing and simulation SDKs to speed Turing’s capabilities to key graphics applications addressing millions of designers, artists and scientists.

Huang also announcing that NVIDIA is open sourcing its Material Definition Language software development kit, starting today.

“We now have a brand new software stack for computer graphics merging rastering and ray tracing, computing and AI,” Huang said.

Bringing Ray-Tracing to Real Time

To put Turing’s capabilities into perspective, Huang opened his keynote talk with a video telling the visual history of computer graphics over the past decades narrated by its pioneering figures — many of whom sat in the audience as the video played. It’s the tale of a grand quest to simulate the world we see all around us, one that has captivated some of the world’s brightest minds for decades.

Turing’s dedicated ray-tracing processors — called RT Cores — accelerate the computation of how light and sound travel in 3D environments. Turing accelerates real-time ray tracing by 25x over the previous Pascal generations. It  can be used for final-frame rendering for film effects at more than 30x the speed of CPUs.

A Familiar Demo, Accelerated by a New GPU

To demonstrate this, Huang showed the audience a demo they’d seen —Epic Games’ stunning Star Wars themed Reflections ray-tracing demo — running on hardware they hadn’t. At the Game Developer Conference in March, Reflections ran on a $70,000 DGX Station equipped with four Volta GPUs. This time the demo ran on a single Turing GPU.

“It turns out it was running on this, one single GPU,” Huang said to wild applause as he playfully blinded the camera by angling the gleaming Quadro RTX 8000’s reflective outer shroud. “This is the world’s first ray-tracing GPU.”

An AI for Beauty

At the same time, the Turing architecture’s Tensor Cores — processors that accelerate deep learning training and inferencing — provide up to 500 trillion tensor operations a second. This, in turn, powers AI-enhanced features — such as denoising, resolution scaling and video re-timing — included in the NVIDIA NGX software development kit.

“At some point you can use AI or some heuristics to figure out what are the missing dots and how should we fill it all in, and it allows us to complete the frame a lot faster than we otherwise could,” Huang said, describing the new deep learning-powered technology stack that enables developers to integrate accelerated, enhanced graphics, photo imaging and video processing into applications with pre-trained networks.

“Nothing is more powerful than using deep learning to do that,” Huang said.

Raster, Faster

Turing also cranks through rasterization — the mainstay of interactive graphics — 6x faster than Pascal, Huang said, detailing how technologies such as variable-rate shading, texture-space shading, and multi-view rendering will provide for more fluid interactivity with large models and scenes and improved VR experiences.

Turning to a tested graphics teaching tool, Huang told the story of how visual effects have progressed by using the Cornell Box — a three-dimensional box inside which various objects are displayed. Huang showed off how Turing uses ray-tracing to deliver complex effects — ranging from diffused reflection to refractions to caustics to global illumination  — with stunning photorealism.

Another showstopper: Huang showed off a video featuring a prototype Porsche — illuminated by lights that played across its undulating curves — celebrating the automakers 70th anniversary. While the photoreal demo looks filmed, it’s entirely generated on a Turing GPU running Epic Games’ Unreal Engine. “For the very first time, NVIDIA RTX is making it possible for us to bring accelerated workflows and acceleration to this market,” Huang said.

Creators looking to tackle such projects will have plenty of tools to choose from. In addition to four powerful Turing-powered graphics cards – $2,300 Quadro RTX 5000, the $6,300  Quadro RTX 6000, and the $10,000 Quadro RTX 8000 — Huang also introduced the RTX server.

Equipped with eight Turing GPUs, it’s designed to slash rendering times from hours to minutes. Four 8 x GPU RTX servers can do the rendering work of 240 dual core servers at 1/4th the cost, using 1/10 the space, and consuming 1/11th the power. “Instead of a shot taking five hours or six hours, it now takes just one hour,” Huang said. “It’s going to completely change how people do film.”

Summing up, Huang described Turing as the “world’s first ray-tracing GPU,” and “the single greatest leap that we have ever  made in one generation.”

A Rousing Cybernetic Strut

Huang ended his talk with a demo that had members of the audience dancing their way out the door. Dubbed Sol, it showed a pair of robotic assistants placing glossy white space-age armor onto a lone figure, each piece finding their place with with a satisfying click.

As the protagonist ascends to a hatch — ray-traced reflections of the futuristic environment all around him gleaming from his suit and visor — the now unsupervised robots begin to dance to the immortal, brass-backed 16 bar blues chord progression of 1977’s “Boogie Shoes” by KC and the Sunshine Band.

Hearing the music, the armored figure returns, cocks his head in surprise, and — to the audience’s delight — demonstrates his own loose-limbed, fluid dance moves.

As the screen fades to black, and then to an image of the new Quadro GPU RTX GPU, the music continues to pump. The message is clear: now it’s your turn to take what you’ve seen and dance.

The post NVIDIA CEO Jensen Huang Unveils Turing, Reinventing Computer Graphics appeared first on The Official NVIDIA Blog.

Counting Craters: Come On, Come On, Look a Little Closer — at Solar System History

The moon has captured humans’ imaginations for thousands of years — but an astronomical number of questions remain about its history, and the history of our solar system.

Some of the answers lie in the craters that pockmark the moon’s surface. And with deep learning, scientists are able to see these craters more clearly than ever before.

Think of crater research as “solar system archaeology,” says astrophysicist Mohamad Ali-Dib, a postdoctoral researcher at the University of Toronto’s Centre for Planetary Sciences.

Studying craters is a particularly useful way to study airless environments like the moon, Mercury and Mars, where topographical features remain relatively unchanged over time due to a lack of weather and geologic activity.

Taking a closer look at these craters provides scientists with important clues about both the history of that crater and the evolution of the solar system. But until now, craters were counted and measured by hand using images or altimetry data from satellites.

“We are aware of grad students whose full-time day job is to take images of Mercury, or the moon or Mars, and count every crater by hand,” Ali-Dib said. “It’s an extremely laborious job.”

There are a few problems with hand-counting, he points out. It takes “an army of grads and undergrads” to do the laborious and difficult work. And it’s a flawed system because research assistants may have different criteria for identifying craters — not to mention a drop in accuracy when they get tired.

So Ali-Dib, along with Ari Silburt and their fellow researchers at the university, developed a neural network that identified thousands of previously undiscovered lunar craters in a matter of hours.

Zooming In on the Moon

Craters come in all shapes and sizes, from ones large enough to hide the state of Vermont to tiny ones just a couple meters across. Mars is variously estimated to have between 300,000 and more than 635,000 craters, while the moon has millions.

Scientists are most interested in figuring out the size distribution: how many craters with a given radius exist on the surface. This distribution tells them how big and numerous the impactors that created the craters were, which is information that the astrophysicists can then link to theories of solar system collisions.

One such theory: giant planet instability. Some scientists theorize that in the early days of the solar system, the orbit of gas giants like Jupiter and Saturn became chaotic for a period of time. If this theory is true, the orbit disruption would have thrown asteroids all around the solar system, leading to brutal collisions. An event like that would leave its trace in the size distribution of craters on an environment like the moon.

By developing a computational way to log lunar craters, scientists can gain a better idea of their size distribution — which, in turn, gives them more data to validate current theories of solar system history.

The researchers used NVIDIA Tesla P100 GPUs on SciNet HPC Consortium’s P8 supercomputer for both training and inference.

When unleashed on images of the moon, the researchers’ convolutional neural network achieved 92 percent accuracy in finding craters that have already been identified — validation of its ability to correctly spot craters. On top of that, the deep learning model spotted 6,000 new craters in just a few hours. That’s nearly double the number that people have manually identified over decades of research.

Most of these, Ali-Dib noted, were the smaller craters the team was hoping to capture with the neural network. These tiny craters are missing from existing datasets because they’re too small and too many to spend expensive human time recording.

lunar craters identified by neural network
Left: A sample image of the moon from the test data. Center: The researchers’ neural network successfully identified craters that were previously hand-coded (blue), as well as thousands of new ones (red). Right: The hand-coded, ground truth data used to evaluate the neural network. Blue circles are craters successfully matched with the researchers’ method, while purple circles are craters missed by the neural network.

To the Moon, and Beyond

It’s not just the moon that could use a closer look like this: similar satellite data for Mercury and Mars exists. Other airless bodies like asteroids, comets and some of the giant planets’ moons could be studied in the future.

The researchers have already looked at Mercury’s craters using a technique called transfer learning: They took their neural network, which is trained on moon data, and used it to analyze images of Mercury.

The team is additionally looking at other features, like crater depth, for their future work. Another parameter of interest to scientists is the age of a crater. But it takes more than satellite data to figure that out, said Ali-Dib. “You need actual rocks for that.”

The post Counting Craters: Come On, Come On, Look a Little Closer — at Solar System History appeared first on The Official NVIDIA Blog.

Tumor Tracking: How a Neural Network Compares Brain MRIs In a Flash

Binge-watching three seasons of “The Office” can make you feel as if your brain has turned into mush. But in actuality, the brain is always pretty mushy and malleable — making neurosurgery even more difficult than it sounds.

To gauge their success, brain surgeons compare MRI scans taken before and after the procedure to determine whether a tumor has been fully removed.

This process takes time, so if an MRI is being taken mid-operation, the doctor must compare scans by eye. But the brain shifts around during surgery, making that task difficult to accomplish, but no less critical.

Finding a faster way to compare MRI scans could help doctors better treat brain tumors. To that end, a group of MIT researchers has come up with a deep learning solution to compare brain MRIs in under a second.

This could help surgeons check operation success in near real-time during the procedure with intraoperative MRI. It could also help oncologists rapidly analyze how a tumor is responding to treatment by comparing a patient’s MRIs taken over several months or years.

When the Pixels Align

Putting two MRI scans together requires a machine learning algorithm to match each pixel in the original 3D scan to its corresponding location in another scan. It’s not easy to do a good job of this — current state-of-the-art algorithms take up to two hours to align brain scans.

That’s too long to be used for an in-surgery setting. And when hospitals or researchers want to analyze thousands or hundreds of thousands of scans to analyze disease patterns, it’s not practical either.

“For each pixel in one image, the traditional algorithms need to find the approximate location in the other image where the anatomical structures are the same,” said Guha Balakrishnan, an MIT postdoctoral researcher and lead author on the study. “It takes a lot of iterations for these algorithms.”

Using a neural network instead speeds up the process by adding in learning. The researchers’ unsupervised algorithm, called VoxelMorph, learns from unlabeled pairs of MRI scans, quickly identifying what brain structures and features look like and matching the images. Using an NVIDIA TITAN X GPU, this inference work takes about a second to align a pair of scans, compared with a minute on a CPU.

The researchers trained the neural network on a diverse dataset of around 7,000 MRI scans from public sources, using a method called atlas-based registration. This process aligns each training image with a single reference MRI scan, an ideal or average image known as the atlas.

The team is working with Massachusetts General Hospital to run retrospective studies on the millions of scans in their database.

“An experiment that would take two days is now done in a few seconds,” said co-author Adrian Dalca, an MIT postdoctoral fellow. “This enables a new world of research where alignment is just a small step.”

The researchers are working to improve their deep learning model’s performance on lower-quality scans that include noise. This is key for scan alignment to work in a clinical setting.

Research datasets consist of nice, clean scans taken of patients who wait a long time in the MRI machine for a high-quality image. But “if someone’s having a stroke, you want the quickest image possible,” Dalca said. “That’s a different quality scan.”

The team will present a new paper this fall at the medical imaging conference MICCAI. Balakrishnan is also developing a variation of their algorithm that uses semi-supervised learning, combining a small amount of labeled data with an otherwise unlabeled training dataset. He found that this model can improve the neural network’s accuracy by 8 percent, pushing its performance above the traditional, slower algorithms.

Besides brain scans, this alignment solution has potential applications for other medical images like heart and lung CT scans or even ultrasounds, which are particularly noisy, Balakrishnan says. “I think to some degree, it’s unbounded.”

The post Tumor Tracking: How a Neural Network Compares Brain MRIs In a Flash appeared first on The Official NVIDIA Blog.

Recruiting 9 to 5, What AI Way to Make a Living

There’s a revolving door for many jobs in the U.S., and it’s been rotating faster lately. For AI recruiting startup Eightfold, which promises lower churn and faster hiring, that spins up opportunity.

Silicon Valley-based Eightfold, which recently raised $18 million, is offering its talent management platform as U.S. unemployment has hit its lowest since the go-go era of 2000.

Stakes are high for companies seeking the best and the brightest amid a sizzling job market. Many companies — especially tech sector ones — have high churn amid much-needed positions unfilled. Eightfold says its software can slash the time it takes to hire by 40 percent.

“The most important asset in the enterprise is the people,” said Eightfold founder and CEO Ashutosh Garg.

The tech industry veteran helped steer the launch of eight product launches at Google and co-founded BloomReach to offer AI-driven retail personalization.

Tech’s Short Job Tenures

Current recruiting and retention methods are holding back people and employers, Garg says.

The problem is acute in the U.S., as on average it takes two months to fill a position at a cost of as much as $20,000. And churn is unnecessarily high, he says.

This is an escalating workforce problem. The employee tenure on average at tech companies such as Apple, Amazon, Twitter, Microsoft and Airbnb has fallen below two years, according to Paysa, a career service. [Disclosure: NVIDIA’s average tenure of employees is more than 5 years.]

The Eightfold Talent Intelligence Platform is used by companies alongside human resources management software and talent tracking systems on areas such as recruiting, retention and diversity.

Machines ‘More Than 2x Better’

Eightfold’s deep neural network was trained on millions of public profiles and continues to ingest massive helpings of data, including public profiles and info on who is getting hired.

The startup’s ongoing training of its recurrent neural networks, which are predictive by design, enables it to continually improve its inferencing capabilities.

“We use a cluster of NVIDIA GPUs to train deep models. These servers process billions of data points to understand career trajectory of people and predict what they will do next. We have over a dozen deep models powering core components of our product,” Garg said.

The platform has processed over 20 million applications, helping companies increase

response rates from candidates by eightfold (thus its name). It has helped reduce screening costs and time to hire by 90 percent.

“It’s more than two times better at matching than a human recruiter,” Garg said.

AI Matchmaker for Job Candidates

For job matching, AI is a natural fit. Job applicants and recruiters alike can upload a resume into a company’s employment listings portal powered by Eightfold’s AI. The platform shoots back in seconds the handful of jobs that are a good match for the particular candidate.

Recruiters can use the tool to search for a wider set of job candidate attributes. For example, it can be used to find star employees who move up more quickly than others in an organization. It can also find those whose GitHub follower count is higher than average, a key metric for technical recruiting.

The software can ingest data on prospects — education, career trajectory, peer rankings, etc. — to do inferencing about how good of a company fit they are for open jobs.

No, Please Don’t Go!

The software is used for retention as well. The employment tenure on average in the U.S. is 4.2 years and a mere 18 months for millennials. For retention, the software directs its inferencing capabilities at employees to determine what they might do next in their career paths.

It looks at many signals, aiming to determine questions such as: Are you likely to switch jobs? Are you an attrition risk? How engaged are you? Are peers switching roles now?

“There’s all kinds of signals that we try to connect,” Garg said. “You try to give that person another opportunity in the company to keep them from leaving.”

Diversity? Blind Screening

Eightfold offers diversity help for recruiters, too. For those seeking talent, the startup’s algorithms can assist to identify which applicants are a good match. Its diversity function also allows recruiters blind screening of candidates shown, helping to remove bias.

Customers AdRoll and DigitalOcean are among those using the startup’s diversity product.

“By bringing AI into the screening, you are able to remove all these barriers — their gender, their ethnicity and so on,” Garg said. “So many of us know how valuable it is to have diversity.”

The post Recruiting 9 to 5, What AI Way to Make a Living appeared first on The Official NVIDIA Blog.

SuperVize Me: What’s the Difference Between Supervised, Unsupervised, Semi-Supervised and Reinforcement Learning?

There are a few different ways to build IKEA furniture. Each will, ideally, lead to a completed couch or chair. But depending on the details, one approach will make more sense than the others.

Got the instruction manual and all the right pieces? Just follow directions. Getting the hang of it? Toss the manual aside and go solo. But misplace the instructions, and it’s up to you to make sense of that pile of wooden dowels and planks.

It’s the same with deep learning. Based on the kind of data available and the research question at hand, a scientist will choose to train an algorithm using a specific learning model.

In a supervised learning model, the algorithm learns on a labeled dataset, providing an answer key that the algorithm can use to evaluate its accuracy on training data. An unsupervised model, in contrast, provides unlabeled data that the algorithm tries to make sense of by extracting features and patterns on its own.

Semi-supervised learning takes a middle ground. It uses a small amount of labeled data bolstering a larger set of unlabeled data. And reinforcement learning trains an algorithm with a reward system, providing feedback when an artificial intelligence agent performs the best action in a particular situation.

Let’s walk through the kinds of datasets and problems that lend themselves to each kind of learning.

What Is Supervised Learning?

If you’re learning a task under supervision, someone is present judging whether you’re getting the right answer. Similarly, in supervised learning, that means having a full set of labeled data while training an algorithm.

Fully labeled means that each example in the training dataset is tagged with the answer the algorithm should come up with on its own. So, a labeled dataset of flower images would tell the model which photos were of roses, daisies and daffodils. When shown a new image, the model compares it to the training examples to predict the correct label.

With supervised machine learning, the algorithm learns from labeled data.

There are two main areas where supervised learning is useful: classification problems and regression problems.

Cat, koala or turtle? A classification algorithm can tell the difference.

Classification problems ask the algorithm to predict a discrete value, identifying the input data as the member of a particular class, or group. In a training dataset of animal images, that would mean each photo was pre-labeled as cat, koala or turtle. The algorithm is then evaluated by how accurately it can correctly classify new images of other koalas and turtles.

On the other hand, regression problems look at continuous data. One use case, linear regression, should sound familiar from algebra class: given a particular x value, what’s the expected value of the y variable?

A more realistic machine learning example is one involving lots of variables, like an algorithm that predicts the price of an apartment in San Francisco based on square footage, location and proximity to public transport.

Supervised learning is, thus, best suited to problems where there is a set of available reference points or a ground truth with which to train the algorithm. But those aren’t always available.

What Is Unsupervised Learning?

Clean, perfectly labeled datasets aren’t easy to come by. And sometimes, researchers are asking the algorithm questions they don’t know the answer to. That’s where unsupervised learning comes in.

In unsupervised learning, a deep learning model is handed a dataset without explicit instructions on what to do with it. The training dataset is a collection of examples without a specific desired outcome or correct answer. The neural network then attempts to automatically find structure in the data by extracting useful features and analyzing its structure.

Unsupervised learning models automatically extract features and find patterns in the data.

Depending on the problem at hand, the unsupervised learning model can organize the data in different ways.

  • Clustering: Without being an expert ornithologist, it’s possible to look at a collection of bird photos and separate them roughly by species, relying on cues like feather color, size or beak shape. That’s how the most common application for unsupervised learning, clustering, works: the deep learning model looks for training data that are similar to each other and groups them together.
  • Anomaly detection: Banks detect fraudulent transactions by looking for unusual patterns in customer’s purchasing behavior. For instance, if the same credit card is used in California and Denmark within the same day, that’s cause for suspicion. Similarly, unsupervised learning can be used to flag outliers in a dataset.
  • Association: Fill an online shopping cart with diapers, applesauce and sippy cups and the site just may recommend that you add a bib and a baby monitor to your order. This is an example of association, where certain features of a data sample correlate with other features. By looking at a couple key attributes of a data point, an unsupervised learning model can predict the other attributes with which they’re commonly associated.
  • Autoencoders: Autoencoders take input data, compress it into a code, then try to recreate the input data from that summarized code. It’s like starting with Moby Dick, creating a SparkNotes version and then trying to rewrite the original story using only SparkNotes for reference. While a neat deep learning trick, there are fewer real-world cases where a simple autocoder is useful. But add a layer of complexity and the possibilities multiply: by using both noisy and clean versions of an image during training, autoencoders can remove noise from visual data like images, video or medical scans to improve picture quality.

Because there is no “ground truth” element to the data, it’s difficult to measure the accuracy of an algorithm trained with unsupervised learning. But there are many research areas where labeled data is elusive, or too expensive, to get. In these cases, giving the deep learning model free rein to find patterns of its own can produce high-quality results.

What Is Semi-Supervised Learning?

Think of it as a happy medium.

Semi-supervised learning is, for the most part, just what it sounds like: a training dataset with both labeled and unlabeled data. This method is particularly useful when extracting relevant features from the data is difficult, and labeling examples is a time-intensive task for experts.

Semi-supervised learning is especially useful for medical images, where a small amount of labeled data can lead to a significant improvement in accuracy.

Common situations for this kind of learning are medical images like CT scans or MRIs. A trained radiologist can go through and label a small subset of scans for tumors or diseases. It would be too time-intensive and costly to manually label all the scans — but the deep learning network can still benefit from the small proportion of labeled data and improve its accuracy compared to a fully unsupervised model.

A popular training method that starts with a fairly small set of labeled data is using general adversarial networks, or GANs.

Imagine two deep learning networks in competition, each trying to outsmart the other. That’s a GAN. One of the networks, called the generator, tries to create new data points that mimic the training data. The other network, the discriminator, pulls in these newly generated data and evaluates whether they are part of the training data or fakes. The networks improve in a positive feedback loop — as the discriminator gets better at separating the fakes from the originals, the generator improves its ability to create convincing fakes.

This is how a GAN works: The discriminator, labeled “D,” is shown images from both the generator, “G,” and from the training dataset. The discriminator is tasked with determining which images are real, and which are fakes from the generator.

What Is Reinforcement Learning?

Video games are full of reinforcement cues. Complete a level and earn a badge. Defeat the bad guy in a certain number of moves and earn a bonus. Step into a trap — game over.

These cues help players learn how to improve their performance for the next game. Without this feedback, they would just take random actions around a game environment in the hopes of advancing to the next level.

Reinforcement learning operates on the same principle — and actually, video games are a common test environment for this kind of research.

In this kind of machine learning, AI agents are attempting to find the optimal way to accomplish a particular goal, or improve performance on a specific task. As the agent takes action that goes toward the goal, it receives a reward. The overall aim: predict the best next step to take to earn the biggest final reward.

To make its choices, the agent relies both on learnings from past feedback and exploration of new tactics that may present a larger payoff. This involves a long-term strategy — just as the best immediate move in a chess game may not help you win in the long run, the agent tries to maximize the cumulative reward.

It’s an iterative process: the more rounds of feedback, the better the agent’s strategy becomes. This technique is especially useful for training robots, which make a series of decisions in tasks like steering an autonomous vehicle or managing inventory in a warehouse.

Just as students in a school, every algorithm learns differently. But with the diversity of approaches available, it’s only a matter of picking the best way to help your neural network learn the ropes.

The post SuperVize Me: What’s the Difference Between Supervised, Unsupervised, Semi-Supervised and Reinforcement Learning? appeared first on The Official NVIDIA Blog.

NVIDIA and NetApp Team to Help Businesses Accelerate AI

For all the focus these days on AI, it’s largely just the world’s largest hyperscalers that have the chops to roll out predictable, scalable deep learning across their organizations.

Their vast budgets and in-house expertise have been required to design systems with the right balance of compute, storage and networking to deliver powerful AI services across a broad base of users.

NetApp ONTAP AI, powered by NVIDIA DGX and NetApp all-flash storage is a blueprint for enterprises wanting to do the same. It helps organizations, both large and small, transform deep learning ambitions into reality. It offers an easy-to-deploy, modular approach for implementing — and scaling — deep learning across their infrastructures. Deployment times get shrunk from months to days.

We’ve worked with NetApp to distill hard-won design insights and best practices into a replicable formula for rolling out an optimal architecture for AI and deep learning. It’s a formula that eliminates the guesswork of designing infrastructure, providing an optimal configuration of GPU computing, storage and networking.

ONTAP AI is backed by a growing roster of trusted NVIDIA and NetApp partners that can help a business get its deep learning infrastructure up and running quickly and cost effectively. And these partners have the AI expertise and enterprise-grade support needed to keep it humming.

This support can extend into a simplified, day-to-day operational experience that will help ensure the ongoing productivity of an enterprise’s deep learning efforts.

For businesses looking to accelerate and simplify their journey into the AI revolution, ONTAP AI is a great way to get there.

Learn more at https://www.netapp.com/us/products/ontap-ai.aspx.

The post NVIDIA and NetApp Team to Help Businesses Accelerate AI appeared first on The Official NVIDIA Blog.

How Subtle Medical Is Using AI to Slash Costs, Risks of Medical Scans

With $2,000, you could take a round trip to Argentina, Hong Kong or your local hospital’s MRI machine.

In dramatic ways, Subtle Medical, a Silicon Valley startup, is using AI to reduce the costs of magnetic resonance imaging scans — in terms of money, time and radiation exposure.

PET scans and MRIs are used to produce detailed images of a patient’s organs and internal structures, which help doctors diagnose medical conditions. PET scans typically take 45 minutes to capture the full human body, while MRIs can take up to three hours. This means that over 35 million patients each year, in the U.S. alone, collectively spend countless hours and billions of dollars on medical scans.

Subtle Medical, an NVIDIA Inception program award winner, aims to improve the inefficiencies of the process. And by changing the process, they hope to drastically improve the productivity of the hospitals and the experiences of the patients.

Greg Zaharchuk and Enhao Gong, Subtle Medical’s founders, met at Stanford University, where Zaharchuk advises Gong in his doctoral research. Their company uses deep learning to improve image quality, which can reduce the duration of MRI and PET scans by three-quarters.

In addition to reducing the length of time people spend inside MRI scanner, their technology can improve medical scans’ safety. Gadolinium, a potentially harmful metal, is deposited in the body during MRI contrast scans, and its alleged side effects have led to lawsuits. Subtle Medical cuts the amount of contrast dose needed by up to 10 fold.

The company uses NVIDIA GPUs and CUDA to train its deep learning model. GPU computing allowed its research team to accelerate their deep learning process from 1-10 minutes per image to one second per image.

X-Ray Vision

While Subtle Medical could drastically reduce the length of an MRI scan, studies found that accuracy also improved — by up to three times.

The company trained its deep learning model using thousands of images from patients through Stanford. Leveraging this work, Zaharchuck and Gong were able to enhance scans that only used 10 percent of the typical radiation dosage, and generated results that matched the quality of scans with full contrast radiation.

These same findings can also be applied to PET scans, which are commonly used for Alzheimer’s disease and cancer diagnosis. The Subtle Medical team managed to produce the same quality of imaging with an AI-enhanced five-minute exam while also reducing radiation dosage up to 200 fold.

There are many practical applications for Subtle Medical’s research: faster scans mean improved productivity and faster diagnosis. Lower doses mean safer conditions for patients. And an improved workflow means better management of patient needs, along with money and time savings and, most importantly, most importantly, more diagnosed and treated patients.

Going forward, the company hopes to expand to clinical trials, where their findings can directly help patients and hospitals. Despite having been founded just a year ago, Subtle Medical expects to apply for FDA clearance this year.

“We want to bring this technology into clinics so that all the patients and hospitals can benefit from this technology,” Gong said. “We want to empower medical imaging with AI infrastructure to make it more accessible.”

The post How Subtle Medical Is Using AI to Slash Costs, Risks of Medical Scans appeared first on The Official NVIDIA Blog.

With a Few Twists and Turns, Rubik’s Cube May Unlock a Breakthrough in AI

An iconic puzzle from the 1980s — the Rubik’s Cube — is being used to bridge a very contemporary gap between deep learning and advanced mathematics.

Pierre Baldi, the computer science professor overseeing the effort by a team of researchers at the University of California, Irvine, described this gap as the greatest conundrum facing AI today.

“People complain that deep learning is a black box, and that they don’t know what the network is doing,” Baldi said. “We could see that the network was learning mathematics.”

The Rubik’s Cube with multicolored faces has, of course, compelled and confounded people since its invention in 1974 by Hungarian sculptor and architecture professor Ernő Rubik.

The research team’s discovery — that a deep learning model can be used to teach a machine how to do math, in this case the algebraic concept known as group theory — is what Baldi called a “small step in the grand challenge of AI.”

Not the Original Goal

That wasn’t what the researchers set out to do. Rather, they were looking to build a deep learning model that could solve the Rubik’s Cube without any human help, much the way earlier models have mastered the games of chess and go.

They did this by teaching it to approach the cube as a child might.

Starting with a solved puzzle, the model first took one move backward before solving it. Then it took two moves back and solved it, then three moves back, and so on. This forced the algorithm to learn a little more on each effort. Baldi likens it to learning golf by starting first with tap-in puts, then moving further away from the hole as accuracy improves.

In a recently published paper detailing the work, the team gave the reinforcement learning algorithm it developed the moniker “autodidactic iteration.” It was able to solve 100 percent of scrambled cubes in an average of 30 moves, or as quickly as the fastest human solvers.

The model was trained on dozens of machines running NVIDA GPUs, most of them TITANs, in conjunction with the CUDA programming model, TensorFlow machine learning framework and Keras neural network API.

Baldi estimates that the GPUs sped things up by a factor of 5-10x, and that there’s no limit to his team’s ability to put more GPUs to good use in furthering its deep learning research.

“We’re starving for GPUs,” he said. “They are essential to this work.”

An Advancement Ripe with Possibilities

Baldi said that the Rubik’s Cube presents a unique deep learning challenge in that it has only one correct configuration and quintillions of incorrect alternatives. And that’s just working with a traditional three-by-three Rubik’s Cube with nine squares on each side.

Solving larger versions of the puzzle represents the next frontier for the team’s work. They’re interested in seeing how the autodidactic iteration approach works with four-by-four and five-by-five cubes. But first the team has to tweak its approach to take on significant added complexity.

“If you slow down by a factor of two, that’s fine,” Baldi said. “But if you slow down to the speed of continental drift, that’s a problem.”

Baldi also sees opportunities to use the approach of starting with the solution to teach the autodidactic iteration model to master other games.

He believes the work has potential applications in other areas of math beyond group theory, especially math above a high school level, which he said AI has struggled with to date.

If Baldi’s team has any say about it, that struggle could soon become a thing of the past. In the meantime, solving the biggest, baddest puzzles will do.

Try your hand at solving a digital Rubik’s Cube, or watch the UCI team’s deep learning algorhithm solve it, at http://deepcube.igb.uci.edu/.

The post With a Few Twists and Turns, Rubik’s Cube May Unlock a Breakthrough in AI appeared first on The Official NVIDIA Blog.

From Nanotech to AI: NVIDIA Intern Expands Horizons With Deep Learning Institute

Sumati Singh came to NVIDIA this summer as an intern to expand on her academic experience in nanotechnology. An introductory AI course at NVIDIA’s Deep Learning Institute got her thinking bigger than she could have imagined.

That course, Fundamentals of Deep Learning for Computer Vision, prompted her to ponder the wide applications of AI across many fields. She said that taking the course is one of those rare instances that offers a “butterfly effect” — the idea that a small change can have big results.

The Deep Learning Institute coursework was an “eye-opening experience,” she said.

Sumati Singh

The University of Waterloo, Ontario, Canada, senior and area native is pursuing a bachelor’s degree in nanotechnology — that’s really small stuff — and works and gets her kicks on printed circuit boards, or PCBs. Her academic emphasis is in electrical engineering.

Despite an enthusiasm for materials science and engineering, she is now open to whatever wave the tech industry might have in store for her next.

Jump on the GPU

“The thing about the Deep Learning Institute that is unique is its access to the GPU. You can actually train on one. Picking up deep learning is like skiing — you have to do it,” said DLI instructor Michael Mendelson.

Singh jumped right in. It was a big leap for a hardware-oriented person, she said, because she had never previously done anything with machine learning and wasn’t at all familiar with AI concepts or the development frameworks. But the DLI session made it easy to get started and learn using NVIDIA’s ready-to-rock DIGITS environment, she said.

The NVIDIA DIGITS setup allows people to get started right away and bypass an AWS account and finding the right machine learning libraries and more to build an AI system from scratch.

It greatly simplifies the process of getting started and actually working in AI without coding.

“Coding is not my area of expertise, so that was great for me” she said of DIGITS and the class material.

Easy-Peasy Course

The introductory DLI course got her up to speed with a lot of basic concepts before it was even lunch time. After the eight-hour, one-day course at NVIDIA’s Santa Clara campus, she was surprised to be able to hold conversations with people about AI and understand a lot of the basic concepts.

“Exposure to something new like this could completely change your career path,” she said.

After her NVIDIA internship, this up-and-coming AI rockstar plans to complete her studies back in Canada. But her sights are set on a career in Silicon Valley’s tech industry.

For now, she’s glad she got the chance to explore the world of AI. Exposure to these courses could completely change your career path, she said.

“They did a great introductory course. Taking the AI course has broadened my horizons,” Singh said.

Keep an eye out for this rising star.

For more information about our deep learning courses, check out the Deep Learning Institute.

The post From Nanotech to AI: NVIDIA Intern Expands Horizons With Deep Learning Institute appeared first on The Official NVIDIA Blog.

Living the Meme: AI as Funny as Humans for Generating Image Captions

It’s possible to get grad-school credit for writing memes. At least if you use deep learning to do it.

Just ask Lawrence Peirson.

The 23-year-old is pursuing a theoretical astrophysics Ph.D. at Stanford, but decided to enroll in a couple AI courses this year. He and classmate E. Meltem Tolunay came up with a neural network that captions memes for a class project, now published in a whitepaper aptly titled “Dank Learning.” (“Dank,” for the uninitiated, is a synonym for “cool.”)

There are lots of examples of training deep learning models to produce literal captions for an image — for example, accurately captioning an image as “man riding a surfboard” or “child with ice cream cone.” With memes, Peirson’s challenge was to see if a neural network could go beyond literal interpretation and create humorous captions.

Though he was initially skeptical that the memes would be funny, Peirson found that the deep learning model produced “some quite interesting and original humor.”

Attaining Deep Meme

The deep learning network captioned this meme, a variation on the popular advice animals template.To collect training data for the deep learning model, Peirson scraped around 400,000 user-generated memes from the website memegenerator.net. The site provides meme templates and allows users to come up with their own captions.

The dataset included around 3,000 base images, each with many different captions. Since the input data was crowdsourced, there was a wide range in quality of meme captions the deep learning model processed.

“With 400k memes, most aren’t going to be that funny, but at least they teach the system what a meme is, what joke is relevant,” he said.

Internet memes have circulated around the web for years, with a strong foothold in websites like Reddit, Facebook, 9GAG and Quick Meme. The most popular can get more than 2 million unique captions created.

Memes often reference pop culture, current events or esoteric bits of a particular internet subculture. (Peirson runs a meme page called “The specific heat capacity of europium at standard temperature and pressure.”)

As a result, they imbibe both the good and bad of digital culture — the paper notes a bias in the training data towards expletive, racist and sexist memes. Peirson sees the need to filter these out in future work, but points out that “it’s a big problem in natural language processing in general,” not one specific to memes.

The deep learning model was programmed in CUDA and used an NVIDIA TITAN Xp GPU. Peirson and Tolunay tried using both unlabeled data and data labeled with the meme title (for example, success kid or trollface), but saw no significant difference in meme quality.

“They’re very funny in a ‘it sort of makes sense, but not really’ way,” Peirson said. “Memes lend themselves to that kind of humor.”

The deep learning network captioned this meme, a variation on the popular advice animals template.

One Does Not Simply Declare a Meme Dank

To evaluate the deep learning model’s success, the collaborators calculated a perplexity score, which checks whether the neural network can identify clear patterns in the data. They calculated this metric for a few hundred memes with preset formats, such as the Boromir meme, which always begins with the phrase “one does not simply.”

But the true test of a meme is whether it’s funny.

In a qualitative survey, Peirson and his co-author presented people with a human-generated and a deep learning-generated meme, side by side. They asked two questions: whether the subject thought the meme was created by a human or computer, and how the subject would rate the meme’s humor.

The data shows the deep learning memes “were fairly indistinguishable from the real memes,” Peirson says.

They also investigated how the neural network would caption images that were not among the templates in the training dataset. In these cases, the algorithm infers patterns from the unknown image based on what it has seen in the training data. Peirson even showed the deep learning system a photo of his own face to test this, with entertaining results.

When Peirson ran the deep learning model on a photo of his own face, this is one of the captions it came up with.

Memes often go viral, and it seems meme-themed whitepapers are no exception. Peirson says he was “extremely surprised” by the media coverage and wide interest in the project. A complementary mobile app, also titled Dank Learning, will soon be available on the App Store.

This project, he says, has given him a new perspective on how powerful memes can be. Millions of users worldwide consume memes daily on social media sites.

Peirson sees the potential for a strong AI to produce memes “at a whim” on current events to influence public opinion — or for advertisers to use memes for brand awareness: “Having this go viral is an incredible way to market.”

The post Living the Meme: AI as Funny as Humans for Generating Image Captions appeared first on The Official NVIDIA Blog.