Machine learning in the cloud: How it can help you right now

.
What is machine learning?
Machine learning is really about the study of algorithms that have the ability to learn through patterns and, based on that, make predictions against patterns of data. It’s a better alternative to leveraging static program instructions and instead making data-driven predictions or decisions that will improve over time without human intervention and additional programming.

Machine learning could be a game-changer for the business.

One of the concerns, as machine learning becomes more affordable through the use of cloud platforms, is that the technology will be misapplied. This already seems to be a pattern, as cloud providers promote machine learning as having wide value. However, that value won’t be realized if machine learning is applied to systems that can’t benefit from making predictions based on patterns found in data.

So what’s the bottom line with machine learning and the cloud? There is actual value there for businesses, if correctly applied. Enterprises looking for applications for this technology may find that, in some cases, machine learning could be a game-changer for the business.

[ Webinar: Stop Random Acts of Cloud (and Overspending) At Your Organization ]

Finding machine-learning use cases
Machine-learning applications have been widely promoted as the ultimate systems builds that can provide better value to enterprises. However, machine learning is best leveraged for specific types of applications that will benefit the most from this technology, such as fraud detection, predictive marketing, machine monitoring (for the Internet of Things), and inventory management.

Keep in mind that not all machine-learning models are the same. They provide different solution patterns. Most cloud providers, including AWS, Google, and Amazon, provide support for three types of predictions. They have different names, but they boil down to three:

Binary prediction
Category prediction
Value prediction
Let’s explore the potential use cases with each of these.

Binary predictions deal with yes-or-no responses. Use cases where it can be helpful include evaluating data in orders that could suggest fraud, or deciding when it might be worthwhile to try to “upsell” products to a customer based on input from a machine learning-enabled recommendation engine.

The types of applications we leverage for these types of predictions are more numerous than the other types of predictions, considering that the responses are much less complex: yes or no. Thus, these types of machine-learning use cases often are found in business processes such as order processing, credit check systems, and engines used to recommend videos, music, or other products to users based upon gathered data and learned responses.

Category prediction means that we’re able to look at a data set and, based upon learned information, place that information into a particular category. This is useful when much different types of data are being analyzed and a category should be applied so that data can be better understood and processed.

For instance, insurance companies place different claims in specific categories, based upon what’s been learned over the years. An example would be to define the likely cause of an accident, even if the information is not a part of the data, such as “alcohol likely involved,” “likely fraudulent,” or “likely weather-related.” The machine-learning system makes these assignments based upon past learning, such as the time of day that the accident occurred, location, type of damage done, age of driver, etc.

Category predictions can work with many different types of applications, such as when we need to place additional meaning around the data and the direct correlation data is not in the existing database. Finance, manufacturing, and retail are all verticals that can use this type of technology.

Value predictions are more complex but also more insightful. They tell you quantitatively about likely outcomes from the data culled, again, from using learning models to find patterns in the data.

Say we want to find out how many units of a product are likely to sell in the next month. It's good information to know, because it allows us to do tighter manufacturing planning and perhaps economize on travel as salespeople follow up on leads.
The idea is to place these types of predictions within systems that can find this information of value, such as planning and financial systems. Also, they can be part of a management dashboard, so those who make critical decisions in the organization are more likely to find this information of value.

Machine learning is best leveraged for specific types of applications, such as fraud detection, predictive marketing, machine monitoring (for the IoT), and inventory management.

Machine learning in the cloud: What's available today?
Many open-source and proprietary machine-learning systems support the types of predictions described above, and they've been around for years. However, the cost of these systems, in terms of hardware and software, was until recently out of reach for most enterprises. Moreover, even if a business could afford it, it typically did not have the machine-learning talent required to design the prediction models or deal with the data science required.

Enter cloud-based machine-learning solutions from the big three public cloud providers: Google, AWS, and Microsoft. They are very different from each other but share some commonality, advantages, and limitations.

Advantages of available ML systems
These systems are cheap to operate. You only have to pay a few dollars an hour, on average, to drive your very own machine-learning application such as the ones outlined above.

Public clouds also provide cheap data storage. You can leverage true databases or storage systems as the input of the data into the machine learning-enabled applications.

Finally, they all provide SDKs (software developer kits) and APIs that allow you to embed machine-learning functionality directly into applications, and they support most programming languages. The real value of machine-learning technology is the use from within applications, because the types of predictions that are made are more operations- and transaction-focused—for instance, the ability to determine in real time if a loan application is likely fraudulent and provide a process to immediately deal with the issue, perhaps allowing the applicant to fix any errors and resubmit.

Disadvantages
The machine-learning systems on particular public clouds are pretty much bound to those clouds. So if you use a machine-learning system on cloud A, then the data storage mechanism on cloud A will typically be natively supported. However, your enterprise database is not supported unless you provide data integration between your on-premises data storage system and those in the cloud.

Thus, the key value for the cloud provider is clear: If you, the customer, are looking to take advantage of the native machine-learning system, then you will probably want to take advantage of the native storage systems and native databases as well. Also, the applications live better on the cloud platform if they frequently talk to the machine-learning models, which, in turn, often talk to the data. Get the hook?

Of course, if you’re looking to move the data, applications, and other processes to the cloud, you’re fine. The machine-learning system can be accessed as a native cloud service. But if you’re working with hybrid- or multi-cloud deployments—and most of us are—then the separation of the data from the machine-learning engine will be problematic in terms of performance, cost, and usability. Clearly, machine learning could be a loss leader that is designed to attach more enterprises to the cloud.

Learning to make systems that learn
Although machine learning is being sold as a shiny new tool, it’s actually technology that’s been evolving for years. Current IT economics allow us to consider the power of AI, and the AI instances of machine learning, to finally provide value to the enterprise.

A few events brought us to this point.

First, the rise of cheap data storage, cloud and non-cloud, that makes the same massive data sets available from the same source.
Second, the power that we have to process the data, both in sheer processing power, storage, and new big data architectures such as Hadoop.
Third, the use of machine learning and other formally expensive services, now provided as cheap cloud services that can be rented for pennies an hour, in some cases.
Still, machine-learning systems need to be created and managed by those who understand machine learning and data-driven decisions. The limitations are not within the technology, but in the limited number of people who can understand and use it. The skills issue will take much longer to solve, but when we do solve it, we’re looking at technology that can be a real game-changer for most businesses.

How are you using machine learning? Has it helped your business? Share your experiences in the comments section.

Topics: Enterprise IT, Cloud
.
https://techbeacon.com/enterprise-it/machine-learning-cloud-how-it-can-help-you-right-now