1. A New Dimension in Predictive Analytics
In today’s economy, all business is becoming data business. In a study conducted by Forrester Consulting, 98 percent of organizations said that analytics are important to driving business priorities, yet fewer than 40 percent of workloads are leveraging advanced analytics or artificial intelligence. Machine learning offers a way companies can extract greater value from their data to increase revenue, gain competitive advantage and cut costs.
Machine learning is a form of predictive analytics that advances organizations
Analytic solutions based on machine learning often operate in real time, adding a new dimension to BI. While old models will continue to supply key reports and analysis to senior decision-makers, real-time analytics brings information to employees “on the front lines” to improve performance hour-by-hour.
In machine learning—a branch of artificial intelligence—systems are “trained” to use specialized algorithms to study, learn and make predictions and recommendations from huge data troves. Predictive models exposed to new data can adapt without human intervention, learning from previous iterations to produce ever more reliable and repeatable decisions and results.
Over time, this iteration makes systems “smarter”, increasingly able to uncover hidden insights, historical relationships
Machine learning is a powerful analytics technology that’s available right now. Many new commercial and open-source solutions for machine learning are available, along with a rich ecosystem for developers. Chances are good your organization is already using the approach somewhere, such as for spam filtering. Applying machine learning and analytics more widely lets you respond more quickly to dynamic situations and get greater value from your fast-growing troves of data.
2. Predictive Analytics is Everywhere
A big reason for the growing popularity of advanced analytics based on machine learning is that it can deliver business benefits in virtually every industry. Wherever large amounts of data and predictive models need regular adjustment, machine learning makes sense.
Providing recommendations for books, films, clothing and dozens of other categories is a familiar example of machine learning in action. But there are many more.
In retail, machine learning and RFID tagging enable greatly improved inventory management. Simply keeping track of an item’s location items is a big challenge, as is matching physical inventory with book inventory. With machine learning, the data used to solve these problems can also improve product placement and influence customer behavior. For example, the system could scan the physical store for out-of-place inventory in order to relocate
When machine learning is combined with linguistic rules, companies can scan social media to determine what customers are saying about their brand and their products. It can even find hidden, underlying patterns that might indicate excitement or frustration with a particular product.
The technology is already playing a crucial role in applications that involve sensors. Machine learning also is essential for self-driving vehicles, where data from multiple sensors must be coordinated in real time in order to ensure safe decisions.
Machine learning can help analyze geographical data to uncover patterns that can more accurately predict the likelihood that a particular site would be the right location for generating wind or solar power.
These are a few of many examples of machine learning in action. It is a proven technique that is delivering valuable results right now.
3. Distinct Competitive Advantage
Machine learning can provide companies with a competitive edge by solving problems and uncovering insights faster and more easily than conventional analytics. It is especially good at delivering value in three types of situations.
The solution to a problem changes over time: Tracking a brand’s reputation via social media is a good example. Demographics of individual platforms shift; new platforms appear. Changes like these create havoc and force constant revisions for marketers using rules-based analytics to hit the right targets with the right messages. In contrast, machine learning models adapt easily, delivering reliable results over time and freeing resources to solve other problems.
The solution varies from situation to situation: In medicine, for instance. a patient’s personal or family history, age, sex, lifestyle, allergies to certain medications and many other factors make every case different. Machine learning can take all these into account to deliver personalized diagnosis and
The solution exceeds human ability: People can recognize many things, like voices, friend’s faces, certain objects, etc. voices, but may not be able to explain why. The problem? Too many variables. By sifting and categorizing many examples, machine learning can objectively learn to recognize and identify specific external variables that, for example, give a voice its character. (pitch, volume, harmonic overtones, etc.)
The competitive advantage comes from developing machines that don't rely on human sensing, description, intervention, or interaction to solve a new class of decisions. This capability opens up new opportunity many fields, including medicine (cancer screening), manufacturing (defect assessment), and transportation (using sound as an additional cue for driving safety).
4. Faster and Less Expensive
Compared with other analytic approaches, machine learning offers several advantages to IT, data scientists, various line of business groups and their organizations.
Machine learning is nimble and flexible with new data. Rules-based systems do well in static situations, but machine learning excels when data is constantly changing or being added. That’s because it eliminates the need to constantly tweak a system or add rules to get the desired results. This saves development time, and greatly reduces the need for major changes.
Personnel costs for machine learning typically are lower over the long run than conventional analytics. At the beginning, of course, companies must hire highly skilled specialists in probability, statistics, machine learning algorithms, AI training methods, among others. But once machine learning is up and running, predictive models can adjust themselves, meaning fewer humans are needed to tweak for accuracy and reliability.
Another advantage is scalability. Machine learning algorithms are built with parallelism in mind and therefore scale better, which ultimately means faster answers to business problems. Systems that rely on human interaction also don’t scale as well. Machine learning minimizes the need to constantly go back to people for decisions.
Finally, machine learning applications can cost less to run than other types of advanced analytics. Many machine learning techniques easily scale to multiple machines instead of a single, expensive high-end platform.
5. Getting Started with Machine Learning
Success in stepping up to machine learning begins with identifying a business problem where the technology can have a clear, measurable impact. Once a suitable project is identified, organizations must deploy specialists and choose an appropriate technique to teach systems how to think and act. These include:
Supervised learning: The system is given example inputs and outputs, then tasked to form general rules of behavior. Example: The recommendation systems of most major brands use supervised learning to boost the relevance of suggestions and increase sales.
Semi-supervised learning: The system is typically given a small amount of labeled data (with the “right answer”) and a much larger amount of unlabeled data. This mode has the same use cases as supervised learning but is less costly due to lower data costs. It is usually the best choice when the input data is expected to change over time, such as with commodity trading, social media or weather-related situations, for example.
Unsupervised learning: Here, the system simply examines the data looking for structure and patterns. This mode can be used to discover patterns that would otherwise go undiscovered, such as in-store buying behavior that could drive changes in product placement to increase sales.
Reinforcement learning: In this approach, the system is placed in an interactive, changing environment, given a task and provided with feedback in the form of “punishments” and “rewards.” This technique has been used with great success to
Regardless of your project, an organization’s advancement to effectively leveraging machine learning in analytics depends on mastering these foundational practices.
6. Intel: Powerful Processors Are Only the Beginning
Intel helps companies put machine learning to work in real-world applications that demand high-speed performance. It does so with a systems approach that includes processors, optimized software
Machine learning requires high computing horsepower. Intel® Xeon® processors provide a scalable baseline, and the Intel® Xeon Phi™ processor is specifically designed for the highly parallel workloads typical of machine learning, as well as machine learning’s memory and fabric (networking) needs. In one Intel test, this processor delivered a 50x reduction in system training time.1 Intel hardware technology also incorporates programmable and fixed accelerators, memory, storage, and networking capabilities.
In addition, Intel offers the software support that enables IT organizations to move from business problem to
- Libraries and languages with building blocks optimized on Intel Xeon processors. These include the Intel® Math Kernel Library (Intel® MKL) and the Intel® Data Analytics Acceleration Library (Intel® DAAL), as well as the Intel Distribution for Python*.
- Optimized frameworks to simplify development, including Apache Spark*, Caffe*, Torch* and TensorFlow*. Intel enables both open-source and commercial software that lets companies take advantage of the latest processors and system features as soon as they are commercially available.
- Software development kits (SDKs), including Intel® Nervana™ technology, TAP
andthe Intel® Deep Learning SDK. This provides a set of application interfaces so the developer can immediately take advantage of the best machine learning algorithms.
When it comes to optimization, Intel takes multiple approaches. Including coaching customers and vendor partners on ways to make their machine learning code run faster on Intel hardware, as well as implementing some learning functions in silicon, which is always faster.
Finally, Intel engineers are in the field constantly, speaking with IT and line-of-business managers to gain insights on how predictive analytics based on machine learning can solve real-world business problems.