Digital Transformation
cancel
Showing results for 
Search instead for 
Did you mean: 

Systems of Action and why "Everything Computes"

mikeshaw747

Over the last ten years, we have all worked to the model that one compute platform can do it all, and that that compute platform should be housed in a central datacenter. If a task was too much for our existing computing resources, we simply added more CPUs, memory or storage.

Digitization, espeically internet of things, and the breakdown of Moore’s law is going to change all of that.

Towards a world where everything learns

Internet of things are sensing our (analog) reality and in doing so, they generate a huge amount of data. For example, just one self-driving car generates four terabytes of data a day.

And because of the rapid advancement of a branch of artificial intelligence called machine learning, we are able to take this mass of data and “figure out what’s going on”.

Machine learning is not like the computing of the past. Up till now, we have told the computer what to do thru our application programs. With machine learning, the computer teaches itself what to do. As a machine learning system gains experience, it gets better at what it does - without any human intervention.

Such machine learning systems are already in existence, but we will see many, many more in the future.

Using machine learning to help with medical diagnosisUsing machine learning to help with medical diagnosis

 Let’s look at some examples:

  • Fraud detection - machine learning system learns our normal spending patterns, thus allowing it to detect an anomaly and probable fraud
  • Fall detection in the home - cameras monitor people who are vulnerable to falls in their homes. A machine learning system learns what a fall looks like
  • Student pastoral care - the digital footprints of students is fed into a machine learning system. The system learns those digital footprint patterns that lead to a student under-performing and thus needing help from their tutor. Such a system is in use in the UK and it’s able to predict that help is needed four to six weeks ahead of when a tutor might notice a problem
  • Assisting in illness diagnosis - Google’s DeepMind group in London is working with the UK National Health Service on a number of projects where they will create a machine learning system that helps doctors to perform diagnoses. Their first project focussed on kidney disease
  • Fault prediction - there are already a number of predictive maintenance systems in use. General Electric’s new locomotives and FlowServe’s oil pumps have such systems. The machine learning system learns the characteristics of the sensor data that precede a fault, thus allowing for proactive repair, before a breakdown occurs
  • Autonomous vehicles - autonomous vehicles probably get more than their fair share of publicity, but they are, essentially machine learning systems on wheels !
  • Autonomous digital marketing - machine learning figures out the mix of digital advertising that gets the best returns and adjusts this mix accordingly. Machine learning is also used to characterise customers and the “customer journeys” that they are on (www.tamr.com)

Machine learning allows us to create “Systems of Action”

What machine learning allows us to do is create what we call a “System of Action”. Inputs come into the portion of the machine learning system figures out what is going on. These inputs might be video feeds from the perimeter fence of Heathrow Airport in London, or the many sensor inputs from a self-driving car, or the sensor inputs from an oil pump, or the sales data from a digital advertising mix, or the sensor inputs from a Malaysian palm field, or the camera and shake data from an elderly person in their home, or the digital footprints from a student, or the digital footprints and detailed scan from a patient.

Based upon what it has learnt in the past, the system infers what is going on - someone is trying to cut the fence at Heathrow airport, this car is heading towards a stop sign, this pump is shaking itself to bits, our online ads in Sweden aren’t working like they do elsewhere, this elderly person has fallen over, this student should be tracking an A but they are taking a path towards a C grade, this patient has early onset kidney disease.

And the system either advises, asks for permission and then takes action, or takes autonomous action - informs the police, stops the car, turns off the pump and sends out the maintenance team, pulls ads for YouTube in Sweden, notifies the ambulance service, notifies the student’s tutor.

We call this a System of Action because it is just that - a system that senses, infers what’s going on then, can take action. This is shown in the diagram below. I'll explain about the "learn" box in a minute.

A System of ActionA System of Action

As I said at the start of this blog, over the past ten years, we have assumed that all data will go to the central data centre and it will be processed there. But in a world of IoT sensors, including video, the amount of data can run into terabytes. So, if the amount of data is huge the transfer latency and the cost of transmission will mean that the “send it to the central data center” model breaks down. We must process the data locally.

Also, if we need to take action quickly, then regardless of the amount of data, we must process it locally also.

In our running examples, predictive maintenance, security of Heathrow airport, the self-driving vehicle and the diagnostic assist probably fall into this category - the data needs to be processed “at the edge”, near to the place where the data is collected.

IDC estimates that 43% of IoT data processing will occur at the edge by the 2019.

Deep Learning and Digital Design require a new compute paradigm

My description of Systems of Action is not complete, however. I said glibly that “the machine learning system learns”. Typically, this learning is done using techniques like deep learning. Techniques like deep learning need to do an enormous amount of computing. And unfortunately for us, we can’t give these systems the compute power they need by simply adding faster processors. We have to change the way architect our computers.

Hewlett Packard Labs starting working on what they believe is required for this next generation of computer. Their architecture is based on the concept of Memory-Driven Computing, where all processes are able to access all data at the some time, at very high speed.

While the learning system does sit at the centre, but it may well not be based on traditional computer architectures.

“One size fits all” no longer applies. “Analyse in the data centre” no longer applies

Because of digitization, because of IoT, and because of machine learning, we are going to have to rethink where we put our compute resources and what type of compute resources they are.

In 43% of IoT-related situations, we will do our “infer” analytics at the edge.

And if we are using analytics techniques like deep learning, we may well need high performance computing, probably using a memory-driven computer.

One size no longer fits all. Edge versus Core. Conventional compute versus high performance compute.One size no longer fits all. Edge versus Core. Conventional compute versus high performance compute.

HPE video on the concept of Everything Computes

HPE has created a one minute video outlining the concept of Everything Computes.

For more discussion on these topics - IoT, machine learning, high performance computing, edge compute, and deep learning, please go to HPE’s Enterprise.nxt.

"Everything Computes" intro video"Everything Computes" intro video

 

 

Mike Shaw
Director Strategic Marketing

twitter.gif@mike_j_shaw
linkedin.gifMike Shaw

0 Kudos
About the Author

mikeshaw747

Mike has been with HPE for 30 years. Half of that time was in research and development, mainly as an architect. The other 15 years has been spent in product management, product marketing, and now, strategic marketing. .

Labels
Events
Jan 30-31, 2018
Online
Expert Days - 2018
Join us online to talk directly with our HPE experts during online Expert Days. Find information here about past, current, and upcoming Expert Days. P...
Read more
28-30 November
Madrid, Spain
Discover 2017 Madrid
Join us for Hewlett Packard Enterprise Discover 2017 Madrid, taking place 28-30 November at the Feria de Madrid Convention Center
Read more
View all