Predictive Models As A Service Vs. Training As A Service: Part One

Erik Marcade

Machine learning is on every CXO’s mind at this time. We’ve all heard and tested many use cases for machine learning, spanning various domains. I would like to focus on one point in this blog: the emergence of two distinct categories of machine learning solutions, depending on the type of problem that needs to be solved.

This week and next, I will describe in detail these approaches, which I call “macro modeling/predictive models as a service” and “micro modeling/training as a service.” In addition, I’ll highlight some of the challenges that technology executives need to be aware of when making investment decisions.

Predictive models as a service—macro modeling

When talking about machine learning, some obvious use cases that come to mind are autonomous vehicles and machine learning-powered translation systems. These could be described as general-purpose systems powered by predictive or machine learning techniques. Let’s see how these are generated and consumed.

Challenge 1

First, we’re talking about one system that has been trained on a very large corpus of data in order to get the proper results in many different situations. For example, Toyota says it needs 8.8 billion miles to create a safe autonomous car. The same is true for image recognition, where public image data sets contain more than 100 million images (and the true internal data sets used by Yahoo and others are much bigger than that).

To build a general-purpose predictive model thus requires a gigantic amount of data that you have the right to use for this purpose.

Challenge 2

How many intelligent general-purpose systems do we need? Today, we have many teams focused on autonomous vehicles, image classification, or even translation systems. But how many of these systems do we need on the planet? If a system to drive autonomous vehicles is efficient enough to beat the competition, you can expect that there will be 10 systems or so to equip all the cars on the planet. We expect these systems to work well in cities, in the countryside, during day and night, and so on. Producers will compete on the price and reliability of the sensors.

The same is true for translation systems or even image recognition systems. This is a winner-take-all market. If we push it to the limit, how many true artificial intelligence systems do we need on the earth?

Challenge 3

As always with predictive and machine learning, it’s almost never a “fire-and-forget” activity. Your systems need to be continuously updated as new data comes in or specific rare situations occur, which means that you need to connect them to continuous feeds of data collection for continuous updates and monitoring.

This continuous improvement feature has cost impacts, of course, that will push the need for continuous learning or incremental learning. This in turn will also be used in order to start from general-purpose predictive models to specific models for specific contexts.

Technical challenges

Of course, the fact that these systems can be transported is important. It’s nice to have such systems available as REST APIs on the cloud. This means that they will be available only within connected environments, which will solve many, but not all, of the use cases. Typically, an autonomous car must be able to run even if there is no connection.

On the shared models, through services, speed, and concurrency is very important, as well as exchanged data security and privacy. These are technical challenges that have been solved in the SAP Leonardo machine-learning foundation based on SAP Cloud Platform with Cloud Foundry.

Finally, we’re talking about very large data volumes and very large computing power, which impact on direct operation costs.> Consider, too, the fact that we can expect more improvement on this financial equation in the future (such as the introduction of ASICs) and even on pure electrical consumption, not to mention the amount of data traffic.

Next week, I’ll discuss training as a service – micro-modeling.

See how you can turn insight into action, make better decisions, and transform your business.

Comments

Erik Marcade

About Erik Marcade

Erik Marcade is vice president of Advanced Analytics Products at SAP.

How Big Data Can Tell You Which Book To Read Next

JP George

If you enjoy reading, but still haven’t foundyour next book to cozy up with, your smartphone might be able to suggest one. Artificial intelligence (AI) is now able to rank literature to predict the next bestseller – a kind of recommendation system, not based on metadata, but on the patterns and themes found in books.

Publishers around the globe are mining all kinds of data, including what’s in the books themselves, in search of the magic formula for evaluating a book’s market potential. With more informed marketing, publishers hope to better target their customers.

Recommending the popular novel

So, how does AI determine what we want to read? It turns out that certain emotional patterns keep us engaged and interested while reading a novel. Kurt Vonnegut first described the curves of emotional plotlines in 1995. Now, with the help of AI sentiment and emotion analysis, such plotlines can be extracted quantitatively. By combining these plotline curves, researchers from the Stanford Literary Lab claim to be able to detect the next blockbuster novel.

Machines think from data

Under the hood of such an AI sits Big Data and machine learning (ML). The concept of Big Data doesn’t just mean lots of data, but also that the data comes from many different data sources and types (e.g., audio, video, images, text, etc.) that are often unstructured (unlike traditional databases with well-defined fields). ML involves statistical algorithms that utilize sets of multi-type, unstructured data to predict class membership. This is possible by either knowing ahead of time which classes exist and training the ML algorithm by example (supervised learning) or letting the algorithm discover the underlying patterns (unsupervised learning).

ML methods include embedded vector space techniques (principal component analysis, K-nearest neighbor, and support vector machine), decision-tree based techniques (classification and regression tree, random forest), gradient and Bayesian-based methods, artificial neural networks (ANN), and others. Many tutorials on machine learning methods can be found here.

ANNs were among the first algorithms to be applied to solve problems in AI, beginning as long ago as the 1940s. For many reasons, their use has waxed and waned over the years, yet interest has recently resurged along with the unprecedented advance of deep learning. This growth in deep learning has lead to what the New York Times calls the great awakening, given Google’s ability to translate text into more than 100 languages.

How AI uncovers sentiment and emotions from text

Imagine automatically extracting the sentiment or emotional impact of a literary work. For a computer to understand a text, what is called natural language processing (NLP), AI algorithms first find a mathematical representation that a machine can understand and that contains maximal information about the text. A simple representation called “bag-of-words” (as the name implies) is a collection of words that appear together, but with no other particular nexus, from which the frequency of word groups could be ascertained. This may provide enough information for classifying themes, but would fail miserably at understanding sentences if word order is important.

Two representations that can quantify information associated with sentence word order are Word2Vec and GloVe. More about NLP representations can be learned from this tutorial, while a tutorial from TensorFlow on Word2vec is found here.

Once sentences are converted to a meaningful representation, a language model is needed that discerns positive emotions from negative emotions. One method would be to use a supervised learning procedure with deep neural networks, as has been done to understand movie reviews. Another way is to allow the deep neural network to discover the emotional patterns by itself. This is the true power behind deep learning: its ability to teach itself, and with more Big Data, to learn more.

Through this process, the ML can understand at text’s major themes (from the word groupings) and emotion. These factors are the fundamental ingredients for an AI application that will recommend a novel.

From creating Animal Farm summaries to discovering who will be the next Danielle Steel, AI is revolutionizing what and how we will read in the future.

For more on using ML to upend the competition, see Why Machine Learning and Why Now?

Comments

JP George

About JP George

JP George grew up in a small town in Washington. After receiving a Master's degree in Public Relations, JP has worked in a variety of positions, from agencies to corporations all across the globe. Experience has made JP an expert in topics relating to leadership, talent management, and organizational business.

Could Governments Run By Artificial Intelligence Be A Good Thing?

Glen Sawyer

Put Skynet from The Terminator movies to the back of your mind for a minute, and stay with me on this one.

Certain political leaders are reminding us of their fragile humanity with increasing frequency these days. Prone to wild acts of emotion and unable to resist the urge to push their personal agenda at the expense of the greater good, it’s enough to make the concept of an AI-controlled government sound utopian by comparison.

I’m not quite naïve enough to think we’re already at a point where our human leaders could be replaced by an all-seeing, all-knowing, all-doing machine, but artificial intelligence and machine learning are becoming ever more tantalizing in their potential to simplify, accelerate, and improve many aspects of society and our lives.

Keeping reality in check

Governments are beginning to realize this. We’re already seeing small crumbs of evidence that they understand how AI can make public services more efficient and citizen-friendly. But these are very early days in discussing and figuring out how such technology could help us enforce laws, organize labor and welfare, and so on, in ways most people would be comfortable with.

And if the Facebook AI story is anything to go by, we’re still pretty spooked by the idea of an intelligence that can “think,” communicate, and potentially make decisions using methods we might not always understand, so a future in which we’re willingly ruled by a digital overlord remains very distant.

What’s more likely – dare I say, inevitable – is that governments will find ways to take advantage of AI in smaller increments, and this will eventually compound to form a political system in which machines are doing most of the “thinking” work.

Keeping humanity in check

Unless you believe the singularity is possible, that “thinking” will remain under the control of a far more streamlined government made up of regular, everyday humans. Our greatest hope is that the AI-run aspects of governance are powerful and transparent enough that those humans can’t get away with the deceit, selfishness, and emotion-based political decisions that plague us today.

That said, it would likely be a very different group of people to today running an AI government. If governments do come to rely heavily on technology, it could be a few technologists at the top of the tree – the ones who understand how it all works – who find themselves wielding immense power. With the likes of Mark Zuckerberg already accruing vast political influence to do with as they please, that’s a worrying prospect.

Keeping AI programmers in check

Thankfully, it won’t happen in the way some are fearing it might. Government decision-making is so complex, with so many interlinked aspects, that no one or small group of technological minds could comprehend and control it entirely. I also don’t believe people generally hold the Silicon Valley view that technology alone can solve everything. What I’m saying is, let’s embrace AI safe in the knowledge that collectively we’ll be able to keep it and its programmers in check.

If we do, we’re opening a whole new world of possibilities in efficient, logical, and honest governance. It’s essential we don’t let the same thing happen with a tech-run government that we’re letting happen with the internet – where power is consolidating into too few hands. That will take a combination of remembering the democratic principles that got us to this point and educating enough people to understand the technology overseeing us. I, for one, am optimistic we can get there. Please tell me I’m not alone.

This story originally appeared on the SAP Community.

Comments

Glen Sawyer

About Glen Sawyer

Glen Sawyer is National Director of IoT Digital Transformation at SAP. 

Diving Deep Into Digital Experiences

Kai Goerlich

 

Google Cardboard VR goggles cost US$8
By 2019, immersive solutions
will be adopted in 20% of enterprise businesses
By 2025, the market for immersive hardware and software technology could be $182 billion
In 2017, Lowe’s launched
Holoroom How To VR DIY clinics

Link to Sources


From Dipping a Toe to Fully Immersed

The first wave of virtual reality (VR) and augmented reality (AR) is here,

using smartphones, glasses, and goggles to place us in the middle of 360-degree digital environments or overlay digital artifacts on the physical world. Prototypes, pilot projects, and first movers have already emerged:

  • Guiding warehouse pickers, cargo loaders, and truck drivers with AR
  • Overlaying constantly updated blueprints, measurements, and other construction data on building sites in real time with AR
  • Building 3D machine prototypes in VR for virtual testing and maintenance planning
  • Exhibiting new appliances and fixtures in a VR mockup of the customer’s home
  • Teaching medicine with AR tools that overlay diagnostics and instructions on patients’ bodies

A Vast Sea of Possibilities

Immersive technologies leapt forward in spring 2017 with the introduction of three new products:

  • Nvidia’s Project Holodeck, which generates shared photorealistic VR environments
  • A cloud-based platform for industrial AR from Lenovo New Vision AR and Wikitude
  • A workspace and headset from Meta that lets users use their hands to interact with AR artifacts

The Truly Digital Workplace

New immersive experiences won’t simply be new tools for existing tasks. They promise to create entirely new ways of working.

VR avatars that look and sound like their owners will soon be able to meet in realistic virtual meeting spaces without requiring users to leave their desks or even their homes. With enough computing power and a smart-enough AI, we could soon let VR avatars act as our proxies while we’re doing other things—and (theoretically) do it well enough that no one can tell the difference.

We’ll need a way to signal when an avatar is being human driven in real time, when it’s on autopilot, and when it’s owned by a bot.


What Is Immersion?

A completely immersive experience that’s indistinguishable from real life is impossible given the current constraints on power, throughput, and battery life.

To make current digital experiences more convincing, we’ll need interactive sensors in objects and materials, more powerful infrastructure to create realistic images, and smarter interfaces to interpret and interact with data.

When everything around us is intelligent and interactive, every environment could have an AR overlay or VR presence, with use cases ranging from gaming to firefighting.

We could see a backlash touting the superiority of the unmediated physical world—but multisensory immersive experiences that we can navigate in 360-degree space will change what we consider “real.”


Download the executive brief Diving Deep Into Digital Experiences.


Read the full article Swimming in the Immersive Digital Experience.

Comments

Kai Goerlich

About Kai Goerlich

Kai Goerlich is the Chief Futurist at SAP Innovation Center network His specialties include Competitive Intelligence, Market Intelligence, Corporate Foresight, Trends, Futuring and ideation. Share your thoughts with Kai on Twitter @KaiGoe.heif Futu

Tags:

Blockchain: Much Ado About Nothing? How Very Wrong!

Juergen Roehricht

Let me start with a quote from McKinsey, that in my view hits the nail right on the head:

“No matter what the context, there’s a strong possibility that blockchain will affect your business. The very big question is when.”

Now, in the industries that I cover in my role as general manager and innovation lead for travel and transportation/cargo, engineering, construction and operations, professional services, and media, I engage with many different digital leaders on a regular basis. We are having visionary conversations about the impact of digital technologies and digital transformation on business models and business processes and the way companies address them. Many topics are at different stages of the hype cycle, but the one that definitely stands out is blockchain as a new enabling technology in the enterprise space.

Just a few weeks ago, a customer said to me: “My board is all about blockchain, but I don’t get what the excitement is about – isn’t this just about Bitcoin and a cryptocurrency?”

I can totally understand his confusion. I’ve been talking to many blockchain experts who know that it will have a big impact on many industries and the related business communities. But even they are uncertain about the where, how, and when, and about the strategy on how to deal with it. The reason is that we often look at it from a technology point of view. This is a common mistake, as the starting point should be the business problem and the business issue or process that you want to solve or create.

In my many interactions with Torsten Zube, vice president and blockchain lead at the SAP Innovation Center Network (ICN) in Potsdam, Germany, he has made it very clear that it’s mandatory to “start by identifying the real business problem and then … figure out how blockchain can add value.” This is the right approach.

What we really need to do is provide guidance for our customers to enable them to bring this into the context of their business in order to understand and define valuable use cases for blockchain. We need to use design thinking or other creative strategies to identify the relevant fields for a particular company. We must work with our customers and review their processes and business models to determine which key blockchain aspects, such as provenance and trust, are crucial elements in their industry. This way, we can identify use cases in which blockchain will benefit their business and make their company more successful.

My highly regarded colleague Ulrich Scholl, who is responsible for externalizing the latest industry innovations, especially blockchain, in our SAP Industries organization, recently said: “These kinds of use cases are often not evident, as blockchain capabilities sometimes provide minor but crucial elements when used in combination with other enabling technologies such as IoT and machine learning.” In one recent and very interesting customer case from the autonomous province of South Tyrol, Italy, blockchain was one of various cloud platform services required to make this scenario happen.

How to identify “blockchainable” processes and business topics (value drivers)

To understand the true value and impact of blockchain, we need to keep in mind that a verified transaction can involve any kind of digital asset such as cryptocurrency, contracts, and records (for instance, assets can be tangible equipment or digital media). While blockchain can be used for many different scenarios, some don’t need blockchain technology because they could be handled by a simple ledger, managed and owned by the company, or have such a large volume of data that a distributed ledger cannot support it. Blockchain would not the right solution for these scenarios.

Here are some common factors that can help identify potential blockchain use cases:

  • Multiparty collaboration: Are many different parties, and not just one, involved in the process or scenario, but one party dominates everything? For example, a company with many parties in the ecosystem that are all connected to it but not in a network or more decentralized structure.
  • Process optimization: Will blockchain massively improve a process that today is performed manually, involves multiple parties, needs to be digitized, and is very cumbersome to manage or be part of?
  • Transparency and auditability: Is it important to offer each party transparency (e.g., on the origin, delivery, geolocation, and hand-overs) and auditable steps? (e.g., How can I be sure that the wine in my bottle really is from Bordeaux?)
  • Risk and fraud minimization: Does it help (or is there a need) to minimize risk and fraud for each party, or at least for most of them in the chain? (e.g., A company might want to know if its goods have suffered any shocks in transit or whether the predefined route was not followed.)

Connecting blockchain with the Internet of Things

This is where blockchain’s value can be increased and automated. Just think about a blockchain that is not just maintained or simply added by a human, but automatically acquires different signals from sensors, such as geolocation, temperature, shock, usage hours, alerts, etc. One that knows when a payment or any kind of money transfer has been made, a delivery has been received or arrived at its destination, or a digital asset has been downloaded from the Internet. The relevant automated actions or signals are then recorded in the distributed ledger/blockchain.

Of course, given the massive amount of data that is created by those sensors, automated signals, and data streams, it is imperative that only the very few pieces of data coming from a signal that are relevant for a specific business process or transaction be stored in a blockchain. By recording non-relevant data in a blockchain, we would soon hit data size and performance issues.

Ideas to ignite thinking in specific industries

  • The digital, “blockchained” physical asset (asset lifecycle management): No matter whether you build, use, or maintain an asset, such as a machine, a piece of equipment, a turbine, or a whole aircraft, a blockchain transaction (genesis block) can be created when the asset is created. The blockchain will contain all the contracts and information for the asset as a whole and its parts. In this scenario, an entry is made in the blockchain every time an asset is: sold; maintained by the producer or owner’s maintenance team; audited by a third-party auditor; has malfunctioning parts; sends or receives information from sensors; meets specific thresholds; has spare parts built in; requires a change to the purpose or the capability of the assets due to age or usage duration; receives (or doesn’t receive) payments; etc.
  • The delivery chain, bill of lading: In today’s world, shipping freight from A to B involves lots of manual steps. For example, a carrier receives a booking from a shipper or forwarder, confirms it, and, before the document cut-off time, receives the shipping instructions describing the content and how the master bill of lading should be created. The carrier creates the original bill of lading and hands it over to the ordering party (the current owner of the cargo). Today, that original paper-based bill of lading is required for the freight (the container) to be picked up at the destination (the port of discharge). Imagine if we could do this as a blockchain transaction and by forwarding a PDF by email. There would be one transaction at the beginning, when the shipping carrier creates the bill of lading. Then there would be look-ups, e.g., by the import and release processing clerk of the shipper at the port of discharge and the new owner of the cargo at the destination. Then another transaction could document that the container had been handed over.

The future

I personally believe in the massive transformative power of blockchain, even though we are just at the very beginning. This transformation will be achieved by looking at larger networks with many participants that all have a nearly equal part in a process. Today, many blockchain ideas still have a more centralistic approach, in which one company has a more prominent role than the (many) others and often is “managing” this blockchain/distributed ledger-supported process/approach.

But think about the delivery scenario today, where goods are shipped from one door or company to another door or company, across many parties in the delivery chain: from the shipper/producer via the third-party logistics service provider and/or freight forwarder; to the companies doing the actual transport, like vessels, trucks, aircraft, trains, cars, ferries, and so on; to the final destination/receiver. And all of this happens across many countries, many borders, many handovers, customs, etc., and involves a lot of paperwork, across all constituents.

“Blockchaining” this will be truly transformational. But it will need all constituents in the process or network to participate, even if they have different interests, and to agree on basic principles and an approach.

As Torsten Zube put it, I am not a “blockchain extremist” nor a denier that believes this is just a hype, but a realist open to embracing a new technology in order to change our processes for our collective benefit.

Turn insight into action, make better decisions, and transform your business. Learn how.

Comments

Juergen Roehricht

About Juergen Roehricht

Juergen Roehricht is General Manager of Services Industries and Innovation Lead of the Middle and Eastern Europe region for SAP. The industries he covers include travel and transportation; professional services; media; and engineering, construction and operations. Besides managing the business in those segments, Juergen is focused on supporting innovation and digital transformation strategies of SAP customers. With more than 20 years of experience in IT, he stays up to date on the leading edge of innovation, pioneering and bringing new technologies to market and providing thought leadership. He has published several articles and books, including Collaborative Business and The Multi-Channel Company.