What’s the Connection Between Big Data and AI?

When people talk about big data, are they simply referring to numbers and metrics?

Yes.

And no.

Technically, big data is simply bits and bytes—literally, a massive amount (petabytes or more) of data. But to dismiss big data as mere ones and zeroes misses the point. Big data may physically be a collection of numbers, but when placed against proper context, those numbers take on a life of their own.

This is particularly true in the realm of artificial intelligence (AI). AI and big data are intrinsically connected; without big data, AI simply couldn’t learn. From the perspective of the team in charge of Oracle’s Cloud Business Group (CBG) Product Marketing, they liken big data to the human experience. On Oracle’s Practical Path To AI podcast episode Connecting the Dots Between Big Data and AI, team members compare the AI learning process to the human experience.

The short version: the human brain ingests countless experiences every moment. Everything that is taken in by senses is technically a piece of information or data—a note of music, a word in a book, a drop of rain, and so on. Infant brains learn from the very beginning they start taking in sensory information, and the more they encounter, the more they are able to assimilate and process, then respond in new and informed ways.

AI works similarly. The more data an AI model encounters, the more intelligent it can become. Over time, as more and more data processes through the AI model, it becomes increasingly significant. In that sense, AI models are trained by big data, just as human brains are trained by the data accumulated through multiple experiences.

And while this may all seem scary at first, there’s a definite public shift toward trusting AI-driven software. This is discussed further by Oracle’s CBG team on the podcast episode, and it all goes back to the idea of human experiences. In the digital realm, people now have the ability to document, review, rank, and track these experiences. This knowledge becomes data points in big data, thus fed into AI models which start validating or invalidating the experiences. With enough of a sample size, a determination can be made based on “a power of collective knowledge” that grows and creates this network.

However, that doesn’t mean that AI is the authority on everything, even with all the data in the world.

To hear more about this topic—and why human judgment is still a very real and very necessary part of, well, everything—listen to the entire podcast episode Connecting the Dots Between Big Data and AI and be sure to visit Oracle’s Big Data site to stay on top of the latest developments in the field of big data.

Guest author Michael Chen is a senior manager, product marketing with Oracle Analytics.

Related:

  • No Related Posts

Taking the Fear Factor Out of AI

For decades, films like Space Odyssey, War Games, Terminator and The Matrix have depicted the future and what it would be like if artificial intelligence (AI) took over the world. Fast forward to 2019 and AI is quickly becoming a reality. The things we only used to see in the movies are improving our daily lives and we often don’t realize it. We’ve been living with AI assistance for quite some time. We use Waze and Google Maps to help us predict traffic patterns and find the shortest driving routes. We let Roomba navigate our homes … READ MORE

Related:

  • No Related Posts

New ‘Experience Zones’ Offer a Fast Route to AI Expertise

New Dell EMC AI Experiences Zones showcase the business benefits of artificial intelligence and provide ready access to the latest Dell EMC AI solutions. Organizations around the world now recognize the opportunity to put artificial intelligence to work to solve pressing business problems. In one sign of this growing AI momentum, a recent IDC report predicts that worldwide spending on AI systems will jump by 44 percent this year, to more than $35 billion.[1] This push into the brave new world of AI isn’t confined to just certain industries. It’s across the board, according to IDC. … READ MORE

Related:

  • No Related Posts

Where Were You When Artificial Intelligence Transformed the Enterprise?

Where were you when artificial intelligence (AI) came online? Remember that science fiction movie where AI takes over in a near dystopian future? The plot revolves around a crazy scientist who accidentally put AI online, only to realize the mistake too late. Soon the machines became the human’s overlords. While these science fiction scenarios are entertaining, they really just stoke fear and add to the confusion to AI. What enterprises should be worried about regarding AI, is understanding how their competition is embracing it to get a leg up. Where were you when your competition put … READ MORE

Related:

  • No Related Posts

Proxysg integrate with DLP

I need a solution

Hi everyone,

I am new proyxsg and dlp user. Please tell me how to  solve below questoin.

I have a proxysg(version 6.7) and DLP(15.5). When I intergrate with both, then health check, the proysg shows “health check has failed”.  Can anyone help me what should I do for successful integration. Thanks.

0

Related:

  • No Related Posts

Simple, Scalable, Containerized Deep Learning using Nauta

Deep learning is hard. Between organizing, cleaning and labeling data, selecting the right neural network topology, picking the right hyperparameters, and then waiting – hoping – that the model produced is accurate enough to put into production. It can seem like an impossible puzzle for your data science team to solve. But the IT aspect of the puzzle is no less complicated, especially when the environment needs to be multi-user and support distributed model training. From choosing an operating system, to installing libraries, frameworks, dependencies, and development platforms, building the infrastructure to support your company’s deep … READ MORE

Related:

  • No Related Posts

Block specific google document on google doc

I need a solution

Hi team,

I am just wondering if we could block one specific google document url https://drive.google.com/file/d/xxxxxxxxxxxxxxxxxxxxxxx/view?usp=drive_web

I do not want to block the whole domain but only that google document url.

I have tried:

1. https://drive.google.com/file/d/xxxxxxxxxxxxxxxxxxxxxxx/view?usp=drive_web

2. drive.google.com:443/file/d/xxxxxxxxxxxxxxxxxxxxxxx/view?usp=drive_web

Those didn’t work.

0

Related:

  • No Related Posts

Scaling NMT with Intel® Xeon® Scalable Processors

With the continuous research interest in the field of machine translation and efficient neural network architecture designs improving translation quality, there’s a great need to improve its time to solution. Training a better performing Neural Machine Translation (NMT) model still takes days to weeks depending on the hardware, size of the training corpus and the model architecture.



Intel® Xeon® Scalable processors provide incredible leap in scalability and over 90% of the Top500 super computers run on Intel. In this article we show some of the training considerations and effectiveness when scaling a NMT model using Intel® Xeon® Scalable processors.



An NMT model reads a source sentence in a language and passes it to an encoder which builds an intermediate representation and the decoder processes the intermediate representation to produce the translated target sentence in another language.

enc-dec-architecture.pngFigure 1: Encoder-decoder architecture



The figure above shows an encoder-decoder architecture. The English source sentence, “Hello! How are you?” is read and processed by the architecture to produce a translated German sentence “Hallo! Wie geht sind Sie?”. Traditionally, Recurrent Neural Network (RNN) were used in encoders and decoders, but other neural network architectures such as Convolutional Neural Network (CNN) and attention mechanismbased models are also used.



Architecture and environment

Transformer model is one of the interesting architectures in the field of NMT, which is built with variants of attention mechanism in the encoder-decoder part there by replacing the traditional RNNs in the architecture. This model was able to achieve state of the art results in English-German and English-French translation tasks.



multi_head_attention.png

Figure 2: Multi-head attention block



The above figure shows the multi-head attention block used in the transformer model. At a high-level, the scaled dot-product attention can be thought as finding the relevant information, values (V) based on Query (Q) and Keys (K) and multi-head attention could be thought as several attention layers in parallel to get distinct aspects of the input.



We use the Tensorflows’ official model implementation of the transformer model and we’ve also added Horovod to perform distributed training. WMT English-German parallel corpus with 4.5M sentences was used to train the model.



The tests described in this article were performed in house on Zenith super computer in Dell EMC HPC and AI Innovation lab. Zenith is Dell EMC PowerEdge C6420-based cluster, consisting of 388 dual socket nodes powered by Intel® Xeon® Scalable Gold 6148 processors and interconnected with an Intel® Omni-path fabric.



System Information

CPU Model

Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz

Operating System

Red Hat Enterprise Linux Server release 7.4 (Maipo)

Tensorflow Version

1.10.1 with Intel® MKL

Horovod Version

0.15.0

MPI

Open MPI 3.1.2

Note: We used a specific Horovod branch to handle sparse gradients. Which is now part of the main branch in their GitHub repository.



Weak scaling, environment variables and TF configurations

When training using CPUs, weak scaling, environment variables and TF configurations play a vital role in improving the throughput of the deep learning model. Setting it optimally can help with additional performance gains.



Below are the suggestions based on our empirical tests when running 4 processes per node for the transformer (big) model on 50 zenith nodes. We found setting these variables on all our experiments seem to improve the throughput and modifying OMP_NUM_THREADS based on the number of processes per node.



Environment Variables:

export OMP_NUM_THREADS=10

export KMP_BLOCKTIME=0

export KMP_AFFINITY=granularity=fine,verbose,compact,1,0



TF Configurations:

intra_op_parallelism_threads=$OMP_NUM_THREADS

inter_op_parallelism_threads=1



Experimenting with weak scaling options allows to find the optimal number of processes run per node such that the model is fit in the memory and performance doesn’t deteriorate. For some reason TensorFlow creates an extra thread. Hence, to avoid oversubscription it’s better to set the OMP_NUM_THREADS to 9, 19 or 39 when training with 4,2,1 process per node respectively. Although we didn’t see it affecting the throughput performance in our experiments but may affect performance in a very large-scale setup.



Performance can be improved by threading. This can be done by setting OMP_NUM_THREADS, such that the product of its value and number of MPI ranks per node equals the number of available CPU cores per node. KMP_AFFINITY environment variable provides a way to control the interface which binds OpenMP threads to physical processing units. KMP_BLOCKTIME, sets the time in milliseconds that a thread should wait after completing a parallel execution before sleep. TF configurations such as intra_op_parallelism_threads and inter_op_parallelism_threads are used to adjust the thread pools there by optimizing the CPU performance.



effect_of_environment_variables_bold.png

Figure 3: Effect of environment variables



The above results show that there’s a 1.67x improvement when environment variables are set correctly.



Faster distributed training

Training a large neural network architecture could be time consuming even to perform rapid prototyping or hyper parameter tuning. Thanks to distributed training and open source frameworks like Horovod which allows to train a model using multiple workers. In our previous blog we showed the effectiveness of training an AI radiologist with distributed deep learning and using Intel® Xeon® Scalable processors. Here, we show how distributed training improves the performance of machine translation task.





scaling_performance_bold.png

Figure 4: Scaling Performance



The above chart shows the throughput of the transformer (big) model when trained using 1 – 100 zenith nodes. We get a near linear performance when scaling up the number of nodes. Based on our tests, which include setting the correct environment variables and optimal number of processes, we see an 79x improvement on 100 Zenith nodes with 2 processes per node compared to the throughput on single node with 4 processes.



Translation Quality

NMT models’ translation quality is measured in terms of BLEU (Bi-Lingual Evaluation Understudy) score. It’s a measure to compute the difference between the human and machine translated output.



In a previous blog post we explained some of the challenges of large-batch training of deep learning models. Here, we experimented using a large global batch size of 402k tokens to determine the models’ performance on English to German task. Most of the hyper parameters were set same as that of transformer (big) model, the model was trained using 50 Zenith nodes with 4 processes per node and 2010 being the local batch size. The learning rate grows linearly for 4000 steps to 0.001 and then follows inverse square root decay.



Case-Insensitive BLEU Case-Sensitive BLEU
TensorFlow Official Benchmark Results 28.9
Our results 29.15 28.56

Note: Case-Sensitive score not reported in the Tensorflow Official Benchmark.



Above table shows our results on the test set (newstest2014) after training the model for around 2.7 days (26000 steps). We can also see a clear improvement in the translation quality compared to the results posted on Tensorflow Official Benchmarks.



Conclusion

In this post we showed how to effectively train an NMT system using Intel® Xeon® Scalable processors. We also showed some of the best practices for setting the environment variables and the corresponding scaling performance. Based on our experiments and also following other research work on NMT to understand some of the important aspects of scaling an NMT system, we were able obtain a better translation quality and to speed up the training process. With growing research interest in the field of neural machine translation, we expect to see much interesting and improved NMT models in the future.

Related:

  • No Related Posts

The Human Factor of Managed Services Drives Hardware ROI

EMC logo


When you go out for a special occasion, you want an elevated experience that delights, that satisfies your hunger for something remarkable. You know that feeling, that excitement of the senses when you’re with a group of friends or family at a restaurant where the food, the presentation, the service, the ambiance all come together in perfect harmony to create a superior experience that lingers long after the car ride home.

That’s the sensation we want to deliver to Dell EMC Managed Services customers. Every time.

What about the other end of the spectrum, where your anticipation of say, a pleasant—or at least uneventful—flight home after a productive work week is fraught with derailments (e.g., delays, lost luggage, no Wi-Fi), leaving you with unmet expectations and undue stress!

Good and bad customer service interactions can affect brand loyalty and purchasing decisions, especially in our world of instant digital gratification. No matter how fantastic the product may be, customers will not often return to the scene of a bad service experience.

Food for Thought: Service is What We Make It

So, what are the pillars of an excellent customer experience, whether it be epicurean, episodic travel or epic IT managed services? According to the National Restaurant Association, the keys to opening a restaurant that offers great customer service is hiring people who love to serve and then training them to:

  • Be prompt — taking care of customers’ needs in a timely manner
  • Be friendly — striving to offer welcoming service and genuine interest
  • Be available — staying visible to put customers at ease, should they need something
  • Be exceptional — going above and beyond to set yourself apart from competition

Recently, I had to make a tight airline connection and a flight attendant came by my seat with connecting gate information written on a napkin. I smiled at the “human” gesture and carried the napkin in my backpack. It saved me the time of stopping at the terminal departure boards or waiting for my app to update.

On another flight, I was surprised to receive a handwritten note thanking me for my patronage.

I keep both of these notes to remind me of what I can do to emphasize the “service” in managed services. Sometimes, it is the humanity of our actions that make the difference, that matter most. They do not need a separate invoice or a change management ticket or other process-oriented machinations. Being able to truly understand customers’ expectations can be the difference between bearing fruit and eating humble pie.

For an excellent Dell EMC Infrastructure Managed Services experience, I would add the pillars of consistency, passion and creativity for what we do. And make them scalable.

Consistency means being able to generate repeatable success. It is consistent quality that builds a relationship of trust that makes us believe we will continue to be delighted and valued.

In most cases, infrastructure managed services are mission-critical for our customers and their customers. When we power our delivery of technologies and services with passion and creativity, we are in effect sharing our individual and collective human potential.

Steak and Sizzle: The Human Potential that Differentiates Us

If our award-winning technologies are the steak, our managed services are the sizzle.

Years ago, I worked on some of the first voice recognition deployments of hardware in datacenters. It was the wave of the future. Think: HAL 9000, the fictional AI (Artificial Intelligence) computer in 2001 A Space Odyssey (minus the villainous behavior!). My team was always pondering if and how we could massively scale our technological innovation.

Scroll forward and today the world is populated with smart phones, smart homes, and here at Dell, smart infrastructure. Dell has cornered the market in so many solution categories.

Our AI-enabled technologies are unparalleled. We’ve worked diligently to automate and orchestrate processes for our standard services while mitigating conceivable human error across converged environments and backup and storage tasks, right?

Machines can out-process us many millions of times over. We cannot match the exponential speed of our electronic counterparts. Competitors might be satisfied to offer these wonderments as stand-alones. We are compelled toward something greater.

If you have attended any of my presentations, you know I discuss and champion what I call “the IMS Experience.” Dell EMC Managed Services is the link between the human factor and the electronic one. We are the bridge, the coupler, the interlock between humans and machines.

We are poised to go above and beyond the spectacular algorithms and ML (machine learning) capabilities running infrastructure: think of it as applying the napkin approach of human potential. We talk to our customers about time to value, ROI and peace of mind. We offer award-winning products, support and patented services to help transform their businesses. And we know the customer experience will always be important, even—especially—with the progress of robotic process automation (RPA), artificial intelligence (AI) and machine learning (ML) combining into that “Dell Technologies Experience.”

It quickly delineates what differentiates us from our competitors. Now, the question becomes can we enhance, expand and scale our human potential amid the push for automation, orchestration, AI and ML to continually improve managed services?

Absolutely.

Pound for Pound: It’s All in How You Scale

Dell EMC IMS is tasked with ensuring that our managed services match our award-winning technology. We leverage AI, ML and other deep learning formats to help customers modernize, breakthrough and compete across dynamic industries, such as science and healthcare.

I say, don’t ask what your hardware can do for you, ask what your hardware can do with the human potential built in! Machines were designed to accelerate the work of humans. And yet, it takes human ingenuity to ensure the repeated success of our brilliant machine counterparts.

When Alan Turing set out to break the Enigma code machine, he did not use smarter humans. He employed thoughtful humans to create a machine to analyze a machine.

This is exactly what our Support and Deploy Services do every day, what my colleagues within Managed Services do every day. Sure, our web pages delineate every aspect of the logistics, the scope, the terms of exactly what we will provide—detailed menus. But it is through what we call “engineered solutions” that our hardware, software, support and services are architected to work together.

Partnering connected technologies and minds accelerate predictive resolutions, new insights and better successes. It is the human factor found in Services that enables our customers to get the most out of their hardware investments. When we capitalize on human agency, we amplify the velocity, the veracity and the possibilities of what our machine counterparts can do.

Summary

Humans create best practices and build lasting relationships, not machines. Our teams of skilled associates add the sagacity, the wisdom, to smart infrastructure. We assess and discern where hardware cannot. Then, we scale it to the speed of machines and apply consistency. It is here, in this sweet spot that we build trust and an excellent customer experience that lingers.

Bon Appetit!

The post The Human Factor of Managed Services Drives Hardware ROI appeared first on InFocus Blog | Dell EMC Services.


Update your feed preferences


   

   


   


   

submit to reddit
   

Related:

  • No Related Posts