I’ve watched a ton of movies over my lifetime but I have always enjoyed the Arnold Schwarzenegger Terminator films. Not only are there a bunch of good one-liners but some of the actual concepts are coming to fruition today. The concept of the machines becoming “self-aware” has always given me chills. The whole idea of a machine—i.e., a system that has the ability to increase its knowledge by the ingestion of data—is simply mind blowing.
I recently spent some time at the IBM Santa Theresa Silicon Valley Labs with a bunch of Ph.D.s and got a firsthand glimpse into the future, which is actually happening right now. Enter IBM Machine Learning.
The one thing that almost all corporations have is an abundance of data. The problem has always been how to utilize that data into workable use cases to bring value to running their businesses better. The growth of data has been staggering
- Over 90 percent of the data in the world today has been created in the last 2 years
- Our current output of data is 2.5 quintillion bytes per day
- The internet population has grown 7.5 percent since 2016 to 3.7 billion humans
- The U.S. alone creates 2,657,700 GB of data every minute
The concept of machine learning has been highlighted as the fourth dimension that really empowers data. Machine learning is the intersection of three things:
1. Customers have things like demand forecasting and personalized offerings
2. Operations have things like process optimization and performance acceleration
3. Issues that make up things like fraud detection and problem prevention
The benefits are cost reduction, revenue increase and the speed of analyzing the mountains of data.
Machine Learning Workflow
The question is, “how does this all work?” The learning process is, as you can imagine, a continuous loop from left to right as the below diagram depicts.
Many professions are involved in this process, but it all starts with the data engineers. The data engineers are the ones that know all about the data as far as source and location.
The next stop in the process is where the mass amounts of data is ingested or, for lack of a better term, read into the system. This is the point where the data scientists get involved to develop mathematical use cases to process the data so that it can be used in the machine learning models. The models are trained so they know how to handle the data and to become better with each ingestion of the data.
Once the learning from the model is in place it is then deployed to production where the actual models become stronger and more knowledgeable. The production engineers are the ones that handle the data and ensure it is useable by the corporate user. As you can infer from the diagram, this is a circular process that makes the models stronger statistically speaking and their percentage of accuracy improves with each execution.
Data is used to train the model. This is typically historical data that has known outcomes and as a result the data can be tested by using subsets of the data. Data scientists will usually hold back a subset of the actual data so that it can be scored and then compared to the actual results from the model as a comparison for accuracy. The data scientist looks for a percentage from the scoring that will provide assurance that the results are valid. The model only gets stronger with more data, hence the circular process of Machine Learning.
From an architecture point of view, what does machine learning actually look like? Working with my previous team of Architects, they always wanted to see juicy architectural diagrams of the solution. To please them and show you what machine learning looks like, view the following diagram.
Deciphering this a bit you can see that there is a big increase in ROI by utilizing a variety of technologies. The data science capabilities have been enhanced by integrating Python and Anaconda. Throw in open industry standards and expanded runtime options and you can see it is a very powerful solution capable of analyzing and learning a ton of solid data.
There’s myriad use cases which can be directly associated to the machine learning solution. Below are just a few.
- Identity management
- Customer retention
- Credit monitoring
- Documentation review
- Product recommendations
- Credit risk
Energy and utilities:
- Appliance efficiency
- Billing forecasting
- Energy demand forecasting
- Maximize energy generation
- Prevent customer churn
- Asset performance
- Increased production capacity
- Quality forecasting
- Quality production forecasting
- Internal defect reduction
- Accelerated price determination
- Improved product flow
Healthcare is an industry that is ripe with solid use cases as well. I wanted to briefly highlight one that involved a healthcare company. The challenge this company had was around helping its patients maintain their blood sugar, cholesterol and blood pressure by encouraging good health habits. By utilizing machine learning, they were able to score diabetes patients at the point of sale based on such things as average blood sugar level, cholesterol, blood pressure and if they took their medications on time. Depending on the patient’s score, their co-pay will vary as they request refills of their medications.
The team constructed a classifications node that was able to predict risk categories of each patient. This solution utilized IBM Machine Learning on z/OS. This company positively impacted its client’s health and lowered their cost, thus improving their customers experience. This has led to higher customer retention as well. This is a good example of the benefits of just one use case for machine learning.
Machine Learning Solutions
I am really fortunate to work with some of the top minds in the industry. According to Rajeev Kamath, IBM Client Technical Professional for Z Analytics, machine learning is a very popular topic.
Kamath says that “machine learning is the modern method of predictive analytics that includes analyzing and discovering patterns in data by using technologies derived from artificial intelligence, multivariate statistics, data mining, pattern recognition and advanced analytics.
While the concept of machine learning has been around for decades, in recent years the advances in computing and data collection capabilities and ability to digitize unstructured data have made it possible to deploy machine learning on a widespread basis.
There are many established solution providers who have excellent machine learning solutions. It’s not unusual to see many of these solutions in a single enterprise providing high value, point solutions.
IBM Machine Learning for z/OS
provides unique benefits that are very much in line with many a company’s financial and operational goals:
- Helps simplify the analytics infrastructure and operations
- Enables real time analytics by reducing latency
- Reduces infrastructure, operational and staffing costs related to data movement
In addition, by embracing open source, it gives clients the ultimate flexibility and freedom. IBM Machine Learning exploits the Apache Spark environment, supports many popular analytics languages such as R and Python, in addition to supporting the Predictive Model Markup Language standard. It can also be relatively easily integrated in to a client’s current infrastructure and analytics workflow. It’s a great solution in the market for simplification of architecture and operations, enablement of real time analytics and cost reduction.”
Machine learning is poised to drastically change the way business will be done in our very near future. This is an outstanding disruptive technology that has the ability to harness the power of data.
I heard a quote just the other day that was fascinating around the topic of big data: “Data is the new oil.” How true is that? The machines are indeed becoming “self aware.”
Visit the IBM Machine Learning Hub and learn more about IBM Machine Learning.
Patrick Stanard is an IBM Business Unit Executive for Z Analytics North America Finance. He’s a 35-year professional in the industry, spanning roles as a systems programmer, developer, manager, adjunct faculty member and director of operations. He has a Bachelor of Science in CIS from Saginaw Valley State University and an MBA from Michigan State University