NVIDIA Launches Metropolis Software Partner Program For Deep Learning Platform

The NVIDIA Metropolis intelligent video analytics platform applies deep learning to video streams for applications such as public safety, traffic management and resource optimization.

NVIDIA announces it has brought together a dozen software partners for its Metropolis Software Partner Program.

The program offers a curated list of applications that makes it easy for systems integrators and hardware vendors to build new products, according to the company.

The NVIDIA Metropolis intelligent video analytics platform applies deep learning to video streams for applications such as public safety, traffic management and resource optimization.

The company gives details of the program in a recent blog post:

Together, we’re taking advantage of the more than 1 billion video cameras that will be in our cities by the year 2020 to solve a dizzying array of problems.

Imagine video-based, automated, real-time control of traffic signals to ease traffic congestion. Or cameras mounted on traffic lights that can help you find parking spaces. Or even using cameras to better target preventative maintenance for roads and bridges. Metropolis enables all of this and more.

This software partner program follows the debut in May of our Metropolis edge-to-cloud platform, with its rich set of hardware and SDKs. To make it into our Metropolis Software Partner Program, partners must have production-ready, field-proven solutions.

NVIDIA says deep learning solutions are fueling a growing array of use cases, such as helping first responders react to emergencies more quickly and delivering more personalized experiences to shoppers.

Below is a list of the company’s inaugural partners and applications:

  • Briefcam: Video synopsis solution for quick video review and search, real-time alerting and quantitative video insights.
  • Deepvision AI: Brand, logo and product identification to perform analytics and provide personalized content to retail customers.
  • Herta Security: Facial recognition solutions for law enforcement in crowded public places.
  • Icetana: Self-learning intelligent analytics to detect anomalies in real time from surveillance videos and quickly alert operators.
  • Ironyun: AI video sear­ch using a natural language interface.
  • Omni AI: Self-learning system for anomaly detection on video, SCADA, cyber, image and analytics data in real time.
  • OpenALPR: Automatic license plate recognition software for smart cities, law enforcement and corporations.
  • Sensen Networks: Real-time parking and speed enforcement solutions for cities. Applications for casinos such as table game and customer analytics.
  • Sensetime: Facial recognition solution for public safety, retail and access control.
  • Visionlabs: Facial recognition solution for cooperative (airports) and non-cooperative (stadiums) scenarios at scale.
  • Vocord: Real-time biometric facial recognition software with very high accuracy.
  • Xjera: Solution for people and vehicle counting for enterprise and commercial applications offering high accuracy, high levels of customization and robust security.

Volume Relocating the Tape For Shentong Robot Education and learning Team Co Ltd (8206.HK) – Money Newsweek

     

<!–

Volume Moving the Tape For Shentong Robot Education Group Co Ltd (8206.HK)

–>

Shares of Shentong Robot Education and learning Team Co Ltd (8206.HK) is moving on volatility today -1.16% or -.005 from the open. The HKSE outlined enterprise saw a new bid of .425 on 384000 quantity.

Investing in the inventory marketplace has traditionally available bigger returns than other varieties of investments. With the bigger prospective for returns, there is also a bigger danger variable. Buyers generally will need to address their personal own danger situation ahead of leaping into the marketplace. Figuring out danger appetite can support when selecting which varieties of shares to buy. Some investors will make a decision that they want to choose a likelihood on selected shares that have the prospective to outperform in the future. Other investors may well opt to enjoy it safe and sound and make a portfolio with minimal danger, staple shares. 

Shentong Robot Education and learning Team Co Ltd’s Williams P.c Vary or 14 day Williams %R at the moment sits at -26.67. The Williams %R oscillates in a variety from to -100. A looking at in between and -20 would level to an overbought situation. A looking at from -80 to -100 would sign an oversold situation. The Williams %R was formulated by Larry Williams. This is a momentum indicator that is the inverse of the Quickly Stochastic Oscillator.

Now, the 14-day ADX for Shentong Robot Education and learning Team Co Ltd (8206.HK) is sitting down at 17.28. Generally talking, an ADX price from -25 would point out an absent or weak craze. A price of 25-50 would assistance a robust craze. A price of 50-75 would detect a really robust craze, and a price of 75-100 would guide to an extremely robust craze. ADX is employed to gauge craze toughness but not craze path. Traders frequently insert the Moreover Directional Indicator (+DI) and Minus Directional Indicator (-DI) to detect the path of a craze.

The RSI, or Relative Power Index, is a widely employed technological momentum indicator that compares rate motion about time. The RSI was created by J. Welles Wilder who was striving to measure regardless of whether or not a inventory was overbought or oversold. The RSI may well be handy for spotting irregular rate activity and volatility. The RSI oscillates on a scale from to 100. The ordinary looking at of a inventory will drop in the variety of 30 to 70. A looking at about 70 would point out that the inventory is overbought, and maybe overvalued. A looking at less than 30 may well point out that the inventory is oversold, and maybe undervalued. After a new check, the 14-day RSIfor Shentong Robot Education and learning Team Co Ltd (8206.HK) is at the moment at 58.61, the 7-day stands at 66.32, and the 3-day is sitting down at 58.95.

As we transfer nearer to the conclude of the 12 months, investors may well be enterprise a portfolio overview. Reviewing trades about the previous 6 months, investors must be ready to see what has worked and what has not. There might be some shares that have outperformed the marketplace, and there might be some underperformers as perfectly. Focusing on what has worked so significantly this 12 months may well support present a clearer image for future moves. Pinpointing what went incorrect can also support the trader see which spots of the portfolio will need advancement. If the inventory marketplace carries on on to reach new heights, investors might be on the lookout to lock in some revenue ahead of producing the future huge trade.


Shentong Robotic Education and learning Group Co Ltd (8206.HK) Relocating 2.38% in Session

     

<!–

Shentong Robot Education Group Co Ltd (8206.HK) Moving 2.38% in Session

–>

Shares of Shentong Robotic Education and learning Group Co Ltd (8206.HK) have observed the needle move 2.38% or .01 in the most modern session. The HKSE listed company saw a modern bid of $.43 on 305000 volume. 

Active traders are ordinarily making an attempt to beat the current market any way they can. When commencing out, traders may well be less than the perception that taking gains in the stock current market is straightforward. Though some may well locate this the case, the bulk will comprehend how challenging it in fact is. With so much media interest focused on the working day to working day happenings in the stock current market, it can be straightforward to come to be distracted by all the sounds. Seeking to time the current market is not often a recipe for achievement. Buyers will most probably finish up underwater without having a focused plan. Generating and sustaining a effectively-balanced portfolio may well acquire some time and effort for the beginner investor to sooner or later execute.

Shentong Robotic Education and learning Group Co Ltd’s Williams Percent Variety or 14 working day Williams %R now sits at -20.00. The Williams %R oscillates in a array from to -100. A studying concerning and -20 would position to an overbought scenario. A studying from -80 to -100 would sign an oversold scenario. The Williams %R was formulated by Larry Williams. This is a momentum indicator that is the inverse of the Fast Stochastic Oscillator.

At present, the 14-working day ADX for Shentong Robotic Education and learning Group Co Ltd (8206.HK) is sitting down at 13.97. Frequently speaking, an ADX worth from -25 would suggest an absent or weak craze. A worth of 25-50 would assist a sturdy craze. A worth of 50-75 would discover a pretty sturdy craze, and a worth of 75-100 would direct to an very sturdy craze. ADX is utilised to gauge craze toughness but not craze direction. Traders often increase the Furthermore Directional Indicator (+DI) and Minus Directional Indicator (-DI) to discover the direction of a craze.

The RSI, or Relative Energy Index, is a greatly utilised technical momentum indicator that compares selling price motion in excess of time. The RSI was developed by J. Welles Wilder who was striving to evaluate no matter whether or not a stock was overbought or oversold. The RSI may well be beneficial for recognizing abnormal selling price activity and volatility. The RSI oscillates on a scale from to 100. The standard studying of a stock will fall in the array of 30 to 70. A studying in excess of 70 would suggest that the stock is overbought, and quite possibly overvalued. A studying less than 30 may well suggest that the stock is oversold, and quite possibly undervalued. After a modern examine, the 14-working day RSIfor Shentong Robotic Education and learning Group Co Ltd (8206.HK) is now at 61.96, the 7-working day stands at 76.54, and the 3-working day is sitting down at 95.08.

Shentong Robot Education and learning Team Co Ltd (8206.HK) Relocating 2.38% in Session

     

<!–

Shentong Robot Education Group Co Ltd (8206.HK) Moving 2.38% in Session

–>

Shares of Shentong Robot Education and learning Team Co Ltd (8206.HK) have seen the needle move 2.38% or .01 in the most new session. The HKSE detailed company saw a new bid of $.43 on 305000 volume. 

Energetic buyers are commonly striving to defeat the industry any way they can. When commencing out, buyers might be underneath the impression that getting earnings in the stock industry is easy. Though some might locate this the circumstance, the greater part will notice how tough it truly is. With so significantly media focus centered on the day to day happenings in the stock industry, it can be easy to turn into distracted by all the sound. Trying to time the industry is almost never a recipe for good results. Traders will most possible end up underwater with no a centered program. Building and keeping a properly-balanced portfolio might choose some time and work for the amateur trader to sooner or later attain.

Shentong Robot Education and learning Team Co Ltd’s Williams Per cent Vary or 14 day Williams %R currently sits at -20.00. The Williams %R oscillates in a variety from to -100. A reading amongst and -20 would stage to an overbought problem. A reading from -80 to -100 would sign an oversold problem. The Williams %R was produced by Larry Williams. This is a momentum indicator that is the inverse of the Speedy Stochastic Oscillator.

Currently, the 14-day ADX for Shentong Robot Education and learning Team Co Ltd (8206.HK) is sitting down at 13.97. Usually talking, an ADX price from -25 would reveal an absent or weak development. A price of 25-50 would support a robust development. A price of 50-75 would determine a quite robust development, and a price of 75-100 would lead to an exceptionally robust development. ADX is made use of to gauge development strength but not development direction. Traders often include the Plus Directional Indicator (+DI) and Minus Directional Indicator (-DI) to determine the direction of a development.

The RSI, or Relative Strength Index, is a broadly made use of technological momentum indicator that compares price movement in excess of time. The RSI was created by J. Welles Wilder who was striving to evaluate no matter whether or not a stock was overbought or oversold. The RSI might be helpful for spotting abnormal price activity and volatility. The RSI oscillates on a scale from to 100. The normal reading of a stock will tumble in the variety of 30 to 70. A reading in excess of 70 would reveal that the stock is overbought, and potentially overvalued. A reading underneath 30 might reveal that the stock is oversold, and potentially undervalued. Soon after a new check out, the 14-day RSIfor Shentong Robot Education and learning Team Co Ltd (8206.HK) is currently at 61.96, the 7-day stands at 76.54, and the 3-day is sitting down at 95.08.

Negative Ichimoku Concentrations For Shentong Robot Education and learning Team Co Ltd (8206.HK) Suggest Bearish Momentum – Economic Newsweek

Shares of Shentong Robot Education and learning Team Co Ltd (8206.HK) just lately touched .415, which areas the stock below the Ichimoku cloud, indicating bearish momentum and a possible sell signal for the equity.  Shares of Shentong Robot Education and learning Team Co Ltd opened the last session at .40, touching a superior of .415 and a lower of .39 , yielding a adjust of 0.01.

Ichimoku Kinko Hyo is a technological trend trading charting process that has been used by Japanese commodity and stock sector traders for a long time and is getting increasing reputation among western stock sector traders, currently being generally referred to as Ichimoku Cloud charts. Ichimoku Kinko Hyo, which translates to “equilibrium at a glance chart”, was produced to let a trader to speedily and simply appraise the trend, momentum, and guidance and resistance levels of an asset, from a solitary chart.

The process does include indicators but these should really under no circumstances be regarded in isolation. It is a visible technological evaluation process and the charts are built to be regarded in their entirety to obtain a point of view on the all round course of the share or index and distinguish larger probability chances from lessen probability types. The Ichimoku components are released in a particular buy since that is how you should really examine or trade the sector. The moment you have verified the trend by recognizing selling price as currently being down below or earlier mentioned the cloud, you can transfer to the shifting averages. The most primary concept of this indicator is that if the selling price is earlier mentioned the cloud, the all round trend is bullish though down below the cloud is bearish, and in the cloud is non-biased or unclear. Lastly, when the selling price is earlier mentioned the cloud, then the prime of the cloud will act as a common guidance stage, and when selling price is down below, the cloud base will act as resistance. But don’t forget the cloud has thickness, and hence resistance does as effectively, which by producing these thicker minimizes the risk of a phony breakout.

Another popular indicator among the technological analysts that can enable to evaluate the energy of sector momentum is the Common Directional Index or ADX. The ADX was established by J. Welles Wilder to enable establish how solid a trend is. In common, a soaring ADX line suggests that an present trend is getting energy. The opposite would be the case for a slipping ADX line. At the time of writing, the 14-day ADX for Shentong Robot Education and learning Team Co Ltd (8206.HK) is standing at 10.29. Quite a few chart analysts believe that that an ADX looking through in excess of 25 would advise a solid trend. A looking through underneath 20 would advise no trend, and a looking through from 20-25 would advise that there is no distinct trend signal.

Shentong Robot Education and learning Team Co Ltd (8206.HK)’s Williams Per cent Selection or 14 day Williams %R is sitting down at -41.67. Commonly, if the worth heads earlier mentioned -20, the stock might be regarded to be overbought. On the flip aspect, if the indicator goes underneath -80, this might signal that the stock is oversold. The RSI, or Relative Power Index, is a generally used technological momentum indicator that compares selling price movement in excess of time. The RSI was established by J. Welles Wilder who was striving to evaluate no matter if or not a stock was overbought or oversold. The RSI might be handy for recognizing irregular selling price activity and volatility. The RSI oscillates on a scale from to 100. The regular looking through of a stock will fall in the range of 30 to 70. A looking through in excess of 70 would reveal that the stock is overbought, and potentially overvalued. A looking through underneath 30 might reveal that the stock is oversold, and potentially undervalued. After a recent look at, the 14-day RSI is at present at 50.65, the 7-day stands at 55.86, and the 3-day is sitting down at 74.08.

Having a glance at a further technological stage, Shentong Robot Education and learning Team Co Ltd (8206.HK) presently has a 14-day Commodity Channel Index (CCI) of -9.33. Commonly, the CCI oscillates earlier mentioned and down below a zero line. Regular oscillations are likely to continue to be in the range of -100 to +100. A CCI looking through of +100 might depict overbought situations, though readings in the vicinity of -100 might reveal oversold territory. Whilst the CCI indicator was produced for commodities, it has turn out to be a well-known software for equity analysis as effectively. Shifting common indicators are used commonly for stock evaluation. Quite a few traders will use a combination of shifting averages with distinct time frames to enable critique stock trend course. One of the far more well-known mixtures is to use the 50-day and 200-day shifting averages. Investors might use the 200-day MA to enable easy out the facts a get a clearer long-time period photo. They might glance to the 50-day or 20-day to get a better grasp of what is heading on with the stock in the in the vicinity of-time period. Presently, the 200-day shifting common is at .46 and the 50-day is .41.


NVIDIA morphs from graphics and gaming to AI and deep learning

Maybe you’ve heard of the x86 central processing unit (CPU) architecture that powers most PCs and servers today. But once upon a time in PC land, Intel made a bundle of cash selling x87 math co-processor chips to accompany the x86 products. These chips excelled at, and accelerated, floating point math operations and helped make PCs much faster at performing certain tasks that were hot and relevant back then, like recalculating spreadsheets.

387, redux
But spreadsheets are old hat now, and math co-processor functionality eventually got integrated into the CPU itself, forcing the math x87 line to dry up. But Artificial Intelligence (AI) has, in a way, brought math co-processors back in vogue, by utilizing graphics processing units (GPUs) in a similar supporting role. As it turns out, the kind of mathematical capabilities required to render high-resolution, high frame-rate graphics are also directly applicable to AI.

Specifically, the work required to train predictive machine learning models, especially those based on neural networks and so-called deep learning, involves analyzing large volumes of data, looking for patterns and building statistically-based heuristics. The more training data used, the more accurate the predictive models become. GPUs are great for this type of work, despite the fact that it’s not really about graphics or video.

That’s why NVIDIA, a company originally focused on GPUs and chipsets for video adapter cards and game consoles is rapidly morphing into an AI company.

Health AI
For example, NVIDIA is now working with the Center for Clinical Data Science (CCDS) in Cambridge, Massachusetts, to employ AI in the service of assisting radiologists in reading and interpreting x-rays, MRIs, CAT scans and the like. The company’s DGX systems, based on its Volta AI architecture, are being used by CCDS radiologists to speed up the process of analyzing medical imagery and finding abnormalities and patterns in them.

CCDS just took delivery of the world’s first NVIDIA DGX-1 supercomputer in December of last year and has already successfully trained machine learning models to do work not only in the sphere of radiology but also in cardiology, ophthalmology, dermatology and psychiatry. CCDS will soon be using a DGX Station — an AI-specialized desktop workstation — for medical AI work as well.

Meanwhile, back at the plant

uav-inspecting-transmission-line.jpg

UAV (drone)-based industrial inspection


Credit: NVIDIA

NVIDIA’s DGX technology is being deployed not just in medicine, but in a variety of industrial contexts. For example, the company has teamed with Avitas Systems, a venture backed by General Electric, in the service of drone-assisted industrial inspection. This work involves the physical inspection of industrial infrastructure, including flare stacks and heated gas plumes.

Drones can perform inspections in conditions that would be lethal to humans; NVIDIA explains that flare stacks must be shut down for days before they become cool enough for a human inspector to approach. Such multi-day shut downs involve huge production costs and drone-based inspection saves on those costs.

Evergreen
But drone-based inspection requires real-time intelligent guidance based on readings picked up by the drones’ sensors (including temperatures encountered and what the drone “sees”). That intelligent guidance is only made possible possible by AI, and that’s where the DGX technology comes in. Interestingly, because of all this real-time processing and given the super-human nature of the work, there’s an element to drone inspection that parallels gaming. That’s a pretty cool connection of old and new.

Here’s another one: Because Avitas Systems is a GE venture, it uses GE Predix, which is a predictive analytics platform that integrates with GE Historian. I’ve written about GE historian technology before, but I did so more than five years ago, when its applications were mostly limited to preventive maintenance. That Predix can now support downstream drone-based inspection shows how useful AI is in its industrial applications…and how much value it’s adding to the more rote data collection that has been in place for quite some time.

Detour or destination?
So NVIDIA, a graphics- and video-focused company founded nearly 25 years ago, is reinventing itself as an AI company in the present tense. That’s a great way to stay relevant, but is it orthogonal? After some pondering, I’ve decided it’s not. Not only is math co-processing common to both disciplines in terms of underlying technology, but both offer future-facing technology that can be aimed at rendering immersive experiences and simulations.

Plus, NVIDIA has corroboration from its competitors in making this pivot. AMD’s in the game too with its Radeon Instinct product, and Intel’s Xeon Phi processors are relevant to machine learning and AI as well. Data, analytics and AI are providing the momentum for almost everything in the computing world. Why shouldn’t the semiconductor companies, who are critical to computing’s infrastructure, align with that trend? It’s just common sense.

Nest Learning Thermostat vs. Nest Thermostat E: What’s the difference?

Trying to decide between the Nest Learning Thermostat and the company’s new Nest Thermostat E? Here’s how the two compare!

Nest recently announced the brand new Nest Thermostat E, a less-expensive smart thermostat with many of the same features as its more expensive sibling. The subtle, all-white design and simplified display of the Thermostat E stand in stark contrast to the flashier, heavier look of the Learning Thermostat. But look and feel are only part of the picture. If you’re trying to decide between Nest’s thermostats, this guide will give you an idea of what each brings to the table.

Design

The Nest Learning Thermostat is shown beside the Nest Thermostat E

Other than pricing, the design is arguably the biggest difference between the Learning Thermostat and the new Thermostat E. Where the Learning Thermostat is meant to be a bold bit of industrial art for your wall, the Thermostat E is meant to blend in, almost disappearing into its surroundings.

The Nest Learning Thermostat comes in four colors — white, copper, black, and stainless steel — and the adjustment ring around the outside of the device is made of metal. The Thermostat E comes in one color — white — and its adjustment ring is made of polycarbonate, though Nest says it’s made to feel like ceramic.

The Nest Learning Thermostat’s 2.08-inch, 480 x 480 resolution display is sharp (229 pixels per inch) and bright. The Thermostat E’s 1.76-inch, 320 x 320 resolution display is layered under a polarized covering that dulls the display, helping it to fade into the background and blurring the display’s resolution (182 pixels per inch).

Functionality

You could be forgiven for thinking the Thermostat E doesn’t pack in all of the same “learning” features present in the Nest Learning Thermostat — the new name seems to suggest one “learns” and the other doesn’t. Luckily, that’s not the case!

Both the Nest Learning Thermostat and the Thermostat E pack in all the same functionality, save for one special feature: Farsight. Farsight uses the far-field sensor built into the Nest Learning Thermostat to detect when you walk by, lighting up to display the temperature, time, or weather.

If the Farsight feature is important to you, you’ll want to get the Learning Thermostat. The Thermostat E — like the Learning Thermostat — has temperature, humidity, occupancy, and ambient light sensors, but it doesn’t have those far-field “Farsight” sensors.

As for the rest of the feature set, both thermostats learn as you use them, both can be controlled using the Nest app, both can adjust the temperature based on your location (i.e. turning down when you leave your home), and both offer scheduling.

Compatibility with furnaces

Nest Thermostats work with a whole bunch of heating and cooling systems, but the Thermostat E can’t boast the impressive compatibility rating of its more expensive sibling. Nest says the Learning Thermostat will work with 95% of heating and cooling systems. The Thermostat E, on the other hand, will work with “most” heating and cooling systems. What does that mean, exactly? It means most folks can expect the Thermostat E to work just fine with their setup. Still, if you’ve got a complicated setup with multiple humidifiers, dehumidifiers, fans, and other accessories, it’s worth running a compatibility check before you decided to buy one of these models.

You can use Nest’s Compatibility Checker to see which model(s) will work in your home.

Price

At long last, the largest differentiator between two otherwise similar products: price!

So what’s the takeaway here? If you can skip the far-field sensing feature and would rather have a thermostat that blends in instead of standing out, you can save about $80 by choosing the Nest Thermostat E. That said, premium materials, multiple color options, and Farsight might be worth the extra money for some. Whichever model you choose, you can count on Nest’s nifty learning features that help you save on energy costs over time.

See Nest Learning Thermostat at Amazon

See Next Thermostat E at Amazon

Which will you choose?

Now that you have a better idea of how the two thermostats compare, which one do you think you’ll choose? If you already own Nest’s current Learning Thermostat, is there anything about the Thermostat E that interests you? Give us a shout in the comments!

Meet Michelangelo: Uber’s Machine Learning Platform

Uber Engineering is committed to developing technologies that create seamless, impactful experiences for our customers. We are increasingly investing in artificial intelligence (AI) and machine learning (ML) to fulfill this vision. At Uber, our contribution to this space is Michelangelo, an internal ML-as-a-service platform that democratizes machine learning and makes scaling AI to meet the needs of business as easy as requesting a ride.

Michelangelo enables internal teams to seamlessly build, deploy, and operate machine learning solutions at Uber’s scale. It is designed to cover the end-to-end ML workflow: manage data, train, evaluate, and deploy models, make predictions, and monitor predictions. The system also supports traditional ML models, time series forecasting, and deep learning.

Michelangelo has been serving production use cases at Uber for about a year and has become the de-facto system for machine learning for our engineers and data scientists, with dozens of teams building and deploying models. In fact, it is deployed across several Uber datacenters, leverages specialized hardware, and serves predictions for the highest loaded online services at the company.

In this article, we introduce Michelangelo, discuss product use cases, and walk through the workflow of this powerful new ML-as-a-service system.

 

Motivation behind Michelangelo

Before Michelangelo, we faced a number of challenges with building and deploying machine learning models at Uber related to the size and scale of our operations. While data scientists were using a wide variety of tools to create predictive models (R, scikit-learn, custom algorithms, etc.), separate engineering teams were also building bespoke one-off systems to use these models in production. As a result, the impact of ML at Uber was limited to what a few data scientists and engineers could build in a short time frame with mostly open source tools.

Specifically, there were no systems in place to build reliable, uniform, and reproducible pipelines for creating and managing training and prediction data at scale. Prior to Michelangelo, it was not possible to train models larger than what would fit on data scientists’ desktop machines, and there was neither a standard place to store the results of training experiments nor an easy way to compare one experiment to another. Most importantly, there was no established path to deploying a model into productionin most cases, the relevant engineering team had to create a custom serving container specific to the project at hand. At the same time, we were starting to see signs of many of the ML anti-patterns documented by Scully et al.

Michelangelo is designed to address these gaps by standardizing the workflows and tools across teams though an end-to-end system that enables users across the company to easily build and operate machine learning systems at scale. Our goal was not only to solve these immediate problems, but also create a system that would grow with the business.

When we began building Michelangelo in mid 2015, we started by addressing the challenges around scalable model training and deployment to production serving containers. Then, we focused on building better systems for managing and sharing feature pipelines. More recently, the focus shifted to developer productivityhow to speed up the path from idea to first production model and the fast iterations that follow.

In the next section, we look at an example application to understand how Michelangelo has been used to build and deploy models to solve specific problems at Uber. While we highlight a specific use case for UberEATS, the platform manages dozens of similar models across the company for a variety of prediction use cases.

 

Use case: UberEATS estimated time of delivery model

UberEATS has several models running on Michelangelo, covering meal delivery time predictions, search rankings, search autocomplete, and restaurant rankings. The delivery time models predict how much time a meal will take to prepare and deliver before the order is issued and then again at each stage of the delivery process.

Figure 1: The UberEATS app hosts an estimated delivery time feature powered by machine learning models built on Michelangelo.

Predicting meal estimated time of delivery (ETD) is not simple. When an UberEATS customer places an order it is sent to the restaurant for processing. The restaurant then needs to acknowledge the order and prepare the meal which will take time depending on the complexity of the order and how busy the restaurant is. When the meal is close to being ready, an Uber delivery-partner is dispatched to pick up the meal. Then, the delivery-partner needs to get to the restaurant, find parking, walk inside to get the food, then walk back to the car, drive to the customer’s location (which depends on route, traffic, and other factors), find parking, and walk to the customer’s door to complete the delivery. The goal is to predict the total duration of this complex multi-stage process, as well as recalculate these time-to-delivery predictions at every step of the process.

On the Michelangelo platform, the UberEATS data scientists use gradient boosted decision tree regression models to predict this end-to-end delivery time. Features for the model include information from the request (e.g., time of day, delivery location), historical features (e.g. average meal prep time for the last seven days), and near-realtime calculated features (e.g., average meal prep time for the last one hour). Models are deployed across Uber’s data centers to Michelangelo model serving containers and are invoked via network requests by the UberEATS microservices. These predictions are displayed to UberEATS customers prior to ordering from a restaurant and as their meal is being prepared and delivered.

 

System architecture

Michelangelo consists of a mix of open source systems and components built in-house. The primary open sourced components used are HDFS, Spark, Samza, Cassandra, MLLib, XGBoost, and TensorFlow. We generally prefer to use mature open source options where possible, and will fork, customize, and contribute back as needed, though we sometimes build systems ourselves when open source solutions are not ideal for our use case.

Michelangelo is built on top of Uber’s data and compute infrastructure, providing a data lake that stores all of Uber’s transactional and logged data, Kafka brokers that aggregate logged messages from all Uber’s services, a Samza streaming compute engine, managed Cassandra clusters, and Uber’s in-house service provisioning and deployment tools.

In the next section, we walk through the layers of the system using the UberEATS ETD models as a case study to illustrate the technical details of Michelangelo.

 

Machine learning workflow

The same general workflow exists across almost all machine learning use cases at Uber regardless of the challenge at hand, including classification and regression, as well as time series forecasting. The workflow is generally implementation-agnostic, so easily expanded to support new algorithm types and frameworks, such as newer deep learning frameworks. It also applies across different deployment modes such as both online and offline (and in-car and in-phone) prediction use cases.

We designed Michelangelo specifically to provide scalable, reliable, reproducible, easy-to-use, and automated tools to address the following six-step workflow:  

  1. Manage data
  2. Train models
  3. Evaluate models
  4. Deploy models
  5. Make predictions
  6. Monitor predictions

Next, we go into detail about how Michelangelo’s architecture facilitates each stage of this workflow.

Manage data

Finding good features is often the hardest part of machine learning and we have found that building and managing data pipelines is typically one of the most costly pieces of a complete machine learning solution.

A platform should provide standard tools for building data pipelines to generate feature and label data sets for training (and re-training) and feature-only data sets for predicting. These tools should have deep integration with the company’s data lake or warehouses and with the company’s online data serving systems. The pipelines need to be scalable and performant,  incorporate integrated monitoring for data flow and data quality, and support both online and offline training and predicting. Ideally, they should also generate the features in a way that is shareable across teams to reduce duplicate work and increase data quality. They should also provide strong guard rails and controls to encourage and empower users to adopt best practices (e.g., making it easy to guarantee that the same data generation/preparation process is used at both training time and prediction time).

The data management components of Michelangelo are divided between online and offline pipelines. Currently, the offline pipelines are used to feed batch model training and batch prediction jobs and the online pipelines feed online, low latency predictions (and in the near future, online learning systems).

In addition, we added a layer of data management, a feature store that allows teams to share, discover, and use a highly curated set of features for their machine learning problems.  We found that many modeling problems at Uber use identical or similar features, and there is substantial value in enabling teams to share features between their own projects and for teams in different organizations to share features with each other.

Figure 2: Data preparation pipelines push data into the Feature Store tables and training data repositories.

Offline

Uber’s transactional and log data flows into an HDFS data lake and is easily accessible via Spark and Hive SQL compute jobs. We provide containers and scheduling to run regular jobs to compute features which can be made private to a project or published to the Feature Store (see below) and shared across teams, while batch jobs run on a schedule or a trigger and are integrated with data quality monitoring tools to quickly detect regressions in the pipelineeither due to local or upstream code or data issues.

Online

Models that are deployed online cannot access data stored in HDFS, and it is often difficult to compute some features in a performant manner directly from the online databases that back Uber’s production services (for instance, it is not possible to directly query the UberEATS order service to compute the average meal prep time for a restaurant over a specific period of time). Instead, we allow features needed for online models to be precomputed and stored in Cassandra where they can be read at low latency at prediction time.

We support two options for computing these online-served features, batch precompute and near-real-time compute, outlined below:

  • Batch precompute. The first option for computing is to conduct bulk precomputing and loading historical features from HDFS into Cassandra on a regular basis. This is simple and efficient, and generally works well for historical features where it is acceptable for the features to only be updated every few hours or once a day. This system guarantees that the same data and batch pipeline is used for both training and serving. UberEATS uses this system for features like a ‘restaurant’s average meal preparation time over the last seven days.’
  • Near-real-time compute. The second option is to publish relevant metrics to Kafka and then run Samza-based streaming compute jobs to generate aggregate features at low latency. These features are then written directly to Cassandra for serving and logged back to HDFS for future training jobs. Like the batch system, near-real-time compute ensures that the same data is used for training and serving. To avoid a cold start, we provide a tool to “backfill” this data and generate training data by running a batch job against historical logs. UberEATS uses this near-realtime pipeline for features like a ‘restaurant’s average meal preparation time over the last one hour.’
Shared feature store

We found great value in building a centralized Feature Store in which teams around Uber can create and manage canonical features to be used by their teams and shared with others. At a high level, it accomplishes two things:  

  1. It allows users to easily add features they have built into a shared feature store, requiring only a small amount of extra metadata (owner, description, SLA, etc.) on top of what would be required for a feature generated for private, project-specific usage.
  2. Once features are in the Feature Store, they are very easy to consume, both online and offline, by referencing a feature’s simple canonical name in the model configuration. Equipped with this information, the system handles joining in the correct HDFS data sets for model training or batch prediction and fetching the right value from Cassandra for online predictions.

At the moment, we have approximately 10,000 features in Feature Store that are used to accelerate machine learning projects, and teams across the company are adding new ones all the time. Features in the Feature Store are automatically calculated and updated daily.

In the future, we intend to explore the possibility of building an automated system to search through Feature Store and identify the most useful and important features for solving a given prediction problem.

Domain specific language for feature selection and transformation

Often the features generated by data pipelines or sent from a client service are not in the proper format for the model, and they may be missing values that need to be filled. Moreover, the model may only need a subset of features provided. In some cases, it may be more useful for the model to transform a timestamp into an hour-of-day or day-of-week to better capture seasonal patterns. In other cases, feature values may need to be normalized (e.g., subtract the mean and divide by standard deviation).

To address these issues, we created a DSL (domain specific language) that modelers use to select, transform, and combine the features that are sent to the model at training and prediction times. The DSL is implemented as sub-set of Scala. It is a pure functional language with a complete set of commonly used functions. With this DSL, we also provide the ability for customer teams to add their own user-defined functions. There are accessor functions that fetch feature values from the current context (data pipeline in the case of an offline model or current request from client in the case of an online model) or from the Feature Store.

It is important to note that the DSL expressions are part of the model configuration and the same expressions are applied at training time and at prediction time to help guarantee that the same final set of features is generated and sent to the model in both cases.

Train models

We currently support offline, large-scale distributed training of decision trees, linear and logistic models, unsupervised models (k-means), time series models, and deep neural networks. We regularly add new algorithms in response to customer need and as they are developed by Uber’s AI Labs and other internal researchers. In addition, we let customer teams add their own model types by providing custom training, evaluation, and serving code. The distributed model training system scales up to handle billions of samples and down to small datasets for quick iterations.

A model configuration specifies the model type, hyper-parameters, data source reference, and feature DSL expressions, as well as compute resource requirements (the number of machines, how much memory, whether or not to use GPUs, etc.). It is used to configure the training job, which is run on a YARN or Mesos cluster.

After the model is trained, performance metrics (e.g., ROC curve and PR curve) are computed and combined into a model evaluation report. At the end of training, the original configuration, the learned parameters, and the evaluation report are saved back to our model repository for analysis and deployment.

In addition to training single models, Michelangelo supports hyper-parameter search for all model types as well as partitioned models. With partitioned models, we automatically partition the training data based on configuration from the user and then train one model per partition, falling back to a parent model when needed (e.g. training one model per city and falling back to a country-level model when an accurate city-level model cannot be achieved).

Training jobs can be configured and managed through a web UI or an API, often via Jupyter notebook. Many teams use the API and workflow tools to schedule regular re-training of their models.

Figure 3: Model training jobs use Feature Store and training data repository data sets to train models and then push them to the model repository.

Evaluate models

Models are often trained as part of a methodical exploration process to identify the set of features, algorithms, and hyper-parameters that create the best model for their problem. Before arriving at the ideal model for a given use case, it is not uncommon to train hundreds of models that do not make the cut. Though not ultimately used in production, the performance of these models guide engineers towards the model configuration that results in the best model performance. Keeping track of these trained models (e.g. who trained them and when, on what data set, with which hyper-parameters, etc.), evaluating them, and comparing them to each other are typically big challenges when dealing with so many models and present opportunities for the platform to add a lot of value.

For every model that is trained in Michelangelo, we store a versioned object in our model repository in Cassandra that contains a record of:

  • Who trained the model
  • Start and end time of the training job
  • Full model configuration (features used, hyper-parameter values, etc.)
  • Reference to training and test data sets
  • Distribution and relative importance of each feature
  • Model accuracy metrics
  • Standard charts and graphs for each model type (e.g. ROC curve, PR curve, and confusion matrix for a binary classifier)
  • Full learned parameters of the model
  • Summary statistics for model visualization

The information is easily available to the user through a web UI and programmatically through an API, both for inspecting the details of an individual model and for comparing one or more models with each other.

Model accuracy report

The model accuracy report for a regression model shows standard accuracy metrics and charts. Classification models would display a different set, as depicted below in Figures 4 and 5:

Figure 4: Regression model reports show regression-related performance metrics.

 

Figure 5: Binary classification performance reports show classification-related performance metrics.

Decision tree visualization

For important model types, we provide sophisticated visualization tools to help modelers understand why a model behaves as it does, as well as to help debug it if necessary. In the case of decision tree models, we let the user browse through each of the individual trees to see their relative importance to the overall model, their split points, the importance of each feature to a particular tree, and the distribution of data at each split, among other variables. The user can specify feature values and the visualization will depict the triggered paths down the decision trees, the prediction per tree, and the overall prediction for the model, as pictured in Figure 6 below: 

Figure 6: Tree models can be explored with powerful tree visualizations.

Feature report

Michelangelo provides a feature report that shows each feature in order of importance to the model along with partial dependence plots and distribution histograms. Selecting two features lets the user understand the feature interactions as a two-way partial dependence diagram, as showcased below:

Figure 7: Features, their impact on the model, and their interactions can be explored though a feature report.

Deploy models

Michelangelo has end-to-end support for managing model deployment via the UI or API and three modes in which a model can be deployed:

  1. Offline deployment. The model is deployed to an offline container and run in a Spark job to generate batch predictions either on demand or on a repeating schedule.
  2. Online deployment. The model is deployed to an online prediction service cluster (generally containing hundreds of machines behind a load balancer) where clients can send individual or batched prediction requests as network RPC calls.
  3. Library deployment. We intend to launch a model that is deployed to a serving container that is embedded as a library in another service and invoked via a Java API. (It is not shown in Figure 8, below, but works similarly to online deployment).

Figure 8: Models from the model repository are deployed to online and offline containers for serving.

In all cases, the required model artifacts (metadata files, model parameter files, and compiled DSL expressions) are packaged in a ZIP archive and copied to the relevant hosts across Uber’s data centers using our standard code deployment infrastructure. The prediction containers automatically load the new models from disk and start handling prediction requests.  

Many teams have automation scripts to schedule regular model retraining and deployment via Michelangelo’s API. In the case of the UberEATS delivery time models, training and deployment are triggered manually by data scientists and engineers through the web UI.

Make predictions

Once models are deployed and loaded by the serving container, they are used to make predictions based on feature data loaded from a data pipeline or directly from a client service. The raw features are passed through the compiled DSL expressions which can modify the raw features and/or fetch additional features from the Feature Store. The final feature vector is constructed and passed to the model for scoring. In the case of online models, the prediction is returned to the client service over the network. In the case of offline models, the predictions are written back to Hive where they can be consumed by downstream batch jobs or accessed by users directly through SQL-based query tools, as depicted below: 

Figure 9: Online and offline prediction services use sets of feature vectors to generate predictions.

Referencing models

More than one model can be deployed at the same time to a given serving container. This allows safe transitions from old models to new models and side-by-side A/B testing of models. At serving time, a model is identified by its UUID and an optional tag (or alias) that is specified during deployment. In the case of an online model, the client service sends the feature vector along with the model UUID or model tag that it wants to use; in the case of a tag, the container will generate the prediction using the model most recently deployed to that tag. In the case of batch models, all deployed models are used to score each batch data set and the prediction records contain the model UUID and optional tag so that consumers can filter as appropriate.

If both models have the same signature (i.e. expect the same set of features) when deploying a new model to replace an old model, users can deploy the new model to the same tag as the old model and the container will start using the new model immediately. This allows customers to update their models without requiring a change in their client code. Users can also deploy the new model using just its UUID and then modify a configuration in the client or intermediate service to gradually switch traffic from the old model UUID to the new one.

For A/B testing of models, users can simply deploy competing models either via UUIDs or tags and then use Uber’s experimentation framework from within the client service to send portions of the traffic to each model and track performance metrics.

Scale and latency

Since machine learning models are stateless and share nothing, they are trivial to scale out, both in online and offline serving modes. In the case of online models, we can simply add more hosts to the prediction service cluster and let the load balancer spread the load. In the case of offline predictions, we can add more Spark executors and let Spark manage the parallelism.

Online serving latency depends on model type and complexity and whether or not the model requires features from the Cassandra feature store. In the case of a model that does not need features from Cassandra, we typically see P95 latency of less than 5 milliseconds (ms). In the case of models that do require features from Cassandra, we typically see P95 latency of less than 10ms. The highest traffic models right now are serving more than 250,000 predictions per second.

Monitor predictions

When a model is trained and evaluated, historical data is always used. To make sure that a model is working well into the future, it is critical to monitor its predictions so as to ensure that the data pipelines are continuing to send accurate data and that production environment has not changed such that the model is no longer accurate.

To address this, Michelangelo can automatically log and optionally hold back a percentage of the predictions that it makes and then later join those predictions to the observed outcomes (or labels) generated by the data pipeline. With this information, we can generate ongoing, live measurements of model accuracy. In the case of a regression model, we publish R-squared/coefficient of determination, root mean square logarithmic error (RMSLE), root mean square error (RMSE), and mean absolute error metrics to Uber’s time series monitoring systems so that users can analyze charts over time and set threshold alerts, as depicted below:

Figure 10: Predictions are sampled and compared to observed outcomes to generate model accuracy metrics.

Management plane, API, and web UI

The last important piece of the system is an API tier. This is the brains of the system. It consists of a management application that serves the web UI and network API and integrations with Uber’s system monitoring and alerting infrastructure. This tier also houses the workflow system that is used to orchestrate the batch data pipelines, training jobs, batch prediction jobs, and the deployment of models both to batch and online containers.

Users of Michelangelo interact directly with these components through the web UI, the REST API, and the monitoring and alerting tools.

 

Building on the Michelangelo platform

In the coming months, we plan to continue scaling and hardening the existing system to support both the growth of our set of customer teams and Uber’s business overall. As the platform layers mature, we plan to invest in higher level tools and services to drive democratization of machine learning and better support the needs of our business:

  • AutoML. This will be a system for automatically searching and discovering model configurations (algorithm, feature sets, hyper-parameter values, etc.) that result in the best performing models for given modeling problems. The system would also automatically build the production data pipelines to generate the features and labels needed to power the models. We have addressed big pieces of this already with our Feature Store, our unified offline and online data pipelines, and hyper-parameter search feature. We plan to accelerate our earlier data science work through AutoML. The system would allow data scientists to specify a set of labels and an objective function, and then would make the most privacy-and security-aware use of Uber’s data to find the best model for the problem. The goal is to amplify data scientist productivity with smart tools that make their job easier.
  • Model visualization. Understanding and debugging models is increasingly important, especially for deep learning. While we have made some important first steps with visualization tools for tree-based models, much more needs to be done to enable data scientists to understand, debug, and tune their models and for users to trust the results.
  • Online learning. Most of Uber’s machine learning models directly affect the Uber product in real time. This means they operate in the complex and ever-changing environment of moving things in the physical world. To keep our models accurate as this environment changes, our models need to change with it. Today, teams are regularly retraining their models in Michelangelo. A full platform solution to this use case involves easily updateable model types, faster training and evaluation architecture and pipelines, automated model validation and deployment, and sophisticated monitoring and alerting systems. Though a big project, early results suggest substantial potential gains from doing online learning right.
  • Distributed deep learning. An increasing number of Uber’s machine learning systems are implementing deep learning technologies. The user workflow of defining and iterating on deep learning models is sufficiently different from the standard workflow such that it needs unique platform support. Deep learning use cases typically handle a larger quantity of data, and different hardware requirements (i.e. GPUs) motivate further investments into distributed learning and a tighter integration with a flexible resource management stack.

If you are interesting in tackling machine learning challenges that push the limits of scale, consider applying for a role on our team!  

Jeremy Hermann is an Engineering Manager and Mike Del Balso is a Product Manager on Uber’s Machine Learning Platform team.

 

NVIDIA’s Processors May Soon Power Wal-Mart’s Deep Learning Push | Business Markets and Stocks News

Recently, analyst Trip Chowdhry of Global Equities Research wrote in an investor note that Wal-Mart Stores (NYSE: WMT) will ramp up its focus on deep neural networks for its OneOps cloud business and that the retailer will tap NVIDIA‘s (NASDAQ: NVDA) graphics processing units (GPUs) to make this happen. 

Deep neural networks are used in artificial intelligence processing to allow computers to understand the relationships between pieces of information without having to be specifically programmed to understand that the information is related. Deep neural networks, and the broader deep learning segment, are part of a growing artificial intelligence market.

Chowdhry thinks the ramp-up of Wal-Mart’s cloud will happen over the next six months and will be “incrementally positive” to NVIDIA’s GPU business. These rumors come after reports surfaced in June that Walmart was asking some of its technology customers to move off of Amazon‘s Web Service (AWS) cloud business. Chowdhry thinks the Wal-Mart cloud, running on NVIDIA’s GPUs, will be one-tenth the size of AWS. If it pans out, this would be a significant move for the retail giant and could bring more GPU sales for NVIDIA.

Image source: Getty Images.

Neither NVIDIA nor Wal-Mart has confirmed any of this yet, but if Wal-Mart is looking to bring more deep learning to its cloud business, it would make a lot of sense for it to use NVIDIA’s processors. NVIDIA already powers some of Amazon, Google, and Facebook‘s deep-learning cloud businesses, and the company expects deep learning to become an even bigger part of its GPU sales in the coming years.

NVIDIA’s deep-learning opportunity

While investors will have to wait and see if a Wal-Mart partnership pans out, there’s still plenty of opportunity for NVIDIA in the deep-learning space already. The company is specifically zeroing in on two segments of deep learning that will benefit from the company’s GPUs: deep-learning training and deep-learning inference.

NVIDIA says deep-learning training is when a computer learns a new capability from existing data, and deep-learning inference is when it can apply the capability it just learned to new data — and both come with loads of potential for the company.

The company is doing more than just projecting lots of opportunity in the deep learning markets, though. It already has a leadership position in the deep-learning training market, and Goldman Sachs analyst Toshiya Hari believes the company’s GPUs are used in 90% of artificial intelligence (AI) training systems right now. 

Additionally, NVIDIA said in its second-quarter fiscal 2018 report that it forged new partnerships with Microsoft, Google, Tencent, IBM, Baidu, and Facebook to help them bring new deep learning and artificial intelligence services online. It also announced plans to train 100,000 developers to use deep learning this year. 

Why deep learning is important to NVIDIA

Aside from NVIDIA’s deep-learning total addressable market, adding more of these customers is important, because the company’s data center revenue segment (which includes GPU sales for deep-learning technologies) is becoming a larger part of the business.

In the second quarter, NVIDIA grew its data center revenue by 175% year over year to $416 million. The data center business segment now accounts for 18.6% of total revenue, up from 10.5% a year ago.

Even if Wal-Mart doesn’t tap NVIDIA’s GPUs for its own deep-learning push, NVIDIA is still forging ahead in this space. I’ve said before that NVIDIA’s data center segment is a growing opportunity for the company — and its investors — and that NVIDIA is continually delivering on this potential. I wouldn’t be surprised to see Wal-Mart sign on with NVIDIA soon, but even if it doesn’t, investors should still be very optimistic about the company’s deep-learning opportunities going forward.

10 stocks we like better than Nvidia

When investing geniuses David and Tom Gardner have a stock tip, it can pay to listen. After all, the newsletter they have run for over a decade, Motley Fool Stock Advisor, has tripled the market.*

David and Tom just revealed what they believe are the 10 best stocks for investors to buy right now… and Nvidia wasn’t one of them! That’s right — they think these 10 stocks are even better buys.

*Stock Advisor returns as of August 1, 2017

Teresa Kersten is an employee of LinkedIn and is a member of The Motley Fool’s board of directors. LinkedIn is owned by Microsoft. Chris Neiger has no position in any of the stocks mentioned. The Motley Fool owns shares of and recommends Amazon, Baidu, Facebook, and Nvidia. The Motley Fool has a disclosure policy.

NVIDIA’s Processors May Soon Power Wal-Mart’s Deep Learning Push — The Motley Fool

Recently, analyst Trip Chowdhry of Global Equities Research wrote in an investor note that Wal-Mart Stores (NYSE:WMT) will ramp up its focus on deep neural networks for its OneOps cloud business and that the retailer will tap NVIDIA‘s (NASDAQ:NVDA) graphics processing units (GPUs) to make this happen. 

Deep neural networks are used in artificial intelligence processing to allow computers to understand the relationships between pieces of information without having to be specifically programmed to understand that the information is related. Deep neural networks, and the broader deep learning segment, are part of a growing artificial intelligence market.

Chowdhry thinks the ramp-up of Wal-Mart’s cloud will happen over the next six months and will be “incrementally positive” to NVIDIA’s GPU business. These rumors come after reports surfaced in June that Walmart was asking some of its technology customers to move off of Amazon‘s Web Service (AWS) cloud business. Chowdhry thinks the Wal-Mart cloud, running on NVIDIA’s GPUs, will be one-tenth the size of AWS. If it pans out, this would be a significant move for the retail giant and could bring more GPU sales for NVIDIA.

Brain in a circle with a blue background and lines around it.

Image source: Getty Images.

Neither NVIDIA nor Wal-Mart has confirmed any of this yet, but if Wal-Mart is looking to bring more deep learning to its cloud business, it would make a lot of sense for it to use NVIDIA’s processors. NVIDIA already powers some of Amazon, Google, and Facebook‘s deep-learning cloud businesses, and the company expects deep learning to become an even bigger part of its GPU sales in the coming years.

NVIDIA’s deep-learning opportunity

While investors will have to wait and see if a Wal-Mart partnership pans out, there’s still plenty of opportunity for NVIDIA in the deep-learning space already. The company is specifically zeroing in on two segments of deep learning that will benefit from the company’s GPUs: deep-learning training and deep-learning inference.

NVIDIA says deep-learning training is when a computer learns a new capability from existing data, and deep-learning inference is when it can apply the capability it just learned to new data — and both come with loads of potential for the company.

The company is doing more than just projecting lots of opportunity in the deep learning markets, though. It already has a leadership position in the deep-learning training market, and Goldman Sachs analyst Toshiya Hari believes the company’s GPUs are used in 90% of artificial intelligence (AI) training systems right now. 

Additionally, NVIDIA said in its second-quarter fiscal 2018 report that it forged new partnerships with Microsoft, Google, Tencent, IBM, Baidu, and Facebook to help them bring new deep learning and artificial intelligence services online. It also announced plans to train 100,000 developers to use deep learning this year. 

Why deep learning is important to NVIDIA

Aside from NVIDIA’s deep-learning total addressable market, adding more of these customers is important, because the company’s data center revenue segment (which includes GPU sales for deep-learning technologies) is becoming a larger part of the business.

In the second quarter, NVIDIA grew its data center revenue by 175% year over year to $416 million. The data center business segment now accounts for 18.6% of total revenue, up from 10.5% a year ago.

Even if Wal-Mart doesn’t tap NVIDIA’s GPUs for its own deep-learning push, NVIDIA is still forging ahead in this space. I’ve said before that NVIDIA’s data center segment is a growing opportunity for the company — and its investors — and that NVIDIA is continually delivering on this potential. I wouldn’t be surprised to see Wal-Mart sign on with NVIDIA soon, but even if it doesn’t, investors should still be very optimistic about the company’s deep-learning opportunities going forward.

Teresa Kersten is an employee of LinkedIn and is a member of The Motley Fool’s board of directors. LinkedIn is owned by Microsoft. Chris Neiger has no position in any of the stocks mentioned. The Motley Fool owns shares of and recommends Amazon, Baidu, Facebook, and Nvidia. The Motley Fool has a disclosure policy.