At a recent meeting of the Oracle Product Architects Community, Stephen Green, head of the Information Retrieval and Machine Learning Group within Oracle Labs, was prepared to give a general overview of his organization’s activities. He got through only his first three slides, and then spent the rest of the hour answering a barrage of questions about machine learning and how this specialized programming technique might be applied to the product areas for which attendees were responsible.
The thing about machine learning, the promise of it, is that it has a huge applicability inside of a company like Oracle.”
–Stephen Green, Head of the Information Retrieval Machine Learning Group at Oracle Labs
The interest reflects an extraordinary surge in experimentation and investment in the technology category known as artificial intelligence (AI), of which machine learning is a building block. AI in the form of self-driving cars, game-show-playing computers, and smartphone personal assistants may have grabbed the public’s attention. But machine learning algorithms, now being incorporated into myriad business applications, promise to help lower costs, raise productivity, detect fraud, make better recommendations, improve business processes, gauge customer sentiment, and even root out IT system problems.
Another sign of AI’s current cachet is the frenetic rate at which businesses, universities, and IT organizations alike are snapping up promising startups and making offers to AI experts and data scientists. Apple, for example, has acquired three AI companies since October 2015, the most recent one a startup that uses the technology to read people’s emotions by analyzing facial expressions. GE is recruiting advanced-degreed engineers to staff an AI and robotics lab the company plans to open at its research headquarters in Niskayuna, New York. And GM is advertising for AI computer scientists to help develop its line of self-driving cars along with related state-of-the-art functions such as speech recognition.
Stephen Green, head of the Information Retrieval Machine Learning Group at Oracle Labs in Burlington, Massachusetts, is a man in demand.
“The thing about machine learning, the promise of it, is that it has a huge applicability inside of a company like Oracle,” he says. “Our job is to find places in Oracle’s business where we think machine learning can have an impact.”
Green, who joined Oracle in 2010, has a PhD in computer science from the University of Toronto. He started using machine learning algorithms when he was studying linguistics. “My specialty has always been language and the understanding thereof,” he says.
Now he schools others in machine learning. “Generally speaking, we work with the product groups to get them to understand how machine learning can improve their products, help them understand how to build the models, help them understand how to incorporate machine learning into their business processes,” Green says.
Machine learning, especially a method known as unsupervised learning (see the “Understand the AI Lingo” sidebar), fits well into applications and services that deal with large volumes of data. That’s why Green’s group has been concentrating recently on the areas of big data and data as a service.
For example, he and his colleagues have been working with engineers involved with Oracle Data Cloud services, which make available to customers large sets of consumer and marketing data, assembled for specific audiences. The technology for aggregating and modeling such large data sets was bolstered by two companies Oracle acquired in the last few years, BlueKai and Datalogix.
“These were companies that were already doing sophisticated machine learning,” Green says. “We’re collaborating with [those groups] to do more advanced things.”
Green also has been working with the Oracle Big Data Discovery team, to improve the data analytics offering by refining its underlying technology. Oracle improved its big data proficiency when it acquired a company called Endeca, which developed technology for analyzing unstructured data. “Endeca was more of a search-and-discovery play,” Green says. “We’ve been working with that group to incorporate machine learning into the systems they’ve been building since.”
Green sees plenty of work ahead. “Oracle does essentially everything related to enterprise IT, and there are so many places where a machine learning approach can improve Oracle’s businesses and products,” he says. “Our goal is to collaborate with the product groups to build interesting things that show how machine learning can improve their products, and then to transfer that to them.
Oracle has been incorporating AI techniques into its applications for years, a strategy that’s expanding rapidly. For instance, Oracle recently acquired a startup to add machine learning capabilities to its data-as-a-service offering that will help marketers with personalization and data analytics by uniting consumers’ identities across their various desktop and mobile devices. “Identification methods are different on every device and across every channel,” says Omar Tawakol, general manager of Oracle Data Cloud. “Solving this can enable marketers to have a significantly more effective dialogue with the consumer and save billions of advertising dollars.”
When Reza B’Far, vice president of product development at Oracle, came to the company in 2007 via the acquisition of startup Logical Apps, one of his first undertakings was to develop a suite of financial applications that use AI techniques for governance, risk, and compliance—to ferret out corporate fraud, for example. B’Far now works on Oracle’s portfolio of human capital management applications as well, employing machine learning for such capabilities as “putting together subject matter experts for short-term projects across different parts of an organization,” he says, as well as “suggesting volunteering opportunities that align well with an employee’s interests—like running a race for charity.”
Oracle Human Capital Management Cloud’s recruiting feature uses machine learning algorithms to help HR departments sort through résumés. In the past, HR software would “simply parse through the résumés looking for keywords—the big advantage being the ability to deal with large amounts of data,” says Mark Bennett, work-life and collaborative products strategy director at Oracle. By using machine learning models, “you get better at classifying candidates, especially in the context of a particular kind of business problem or a particular skill, with matches based on the overall semantics of the résumé,” he says.Historical Trends, Heightened Interest
A couple of familiar trends help explain the heightened interest in using machine learning in enterprise applications. Steady increases in compute power and storage laid the groundwork, aided by the introduction of graphics processing units (GPUs), number-crunching chips built for video games that turned out to be especially suited to machine learning models.
Tracking that expansion in power and storage, the volume of data that companies collect has soared, and machine learning algorithms feed on that data. It’s what they use to learn, to figure out patterns, and to spot trends. The more data they have, the more cogent and insightful their output will be.
Oracle is applying machine learning techniques to “every aspect” of Oracle Management Cloud, says Amit Ganesh, senior vice president, Oracle Enterprise Manager. Oracle Management Cloud is a suite of IT infrastructure monitoring, management, and analytics applications delivered as a service. “Machine learning techniques shift the burden from the user asking the right questions to these applications finding the right context-sensitive answers that the user needs to know,” Ganesh says.
Artificial intelligence. The ability of a machine to execute a task without its being programmed specifically for that task. AI is now closely associated with robotics and the ability for a machine to perform human-like tasks, such as image recognition and natural language processing.
Machine learning. An algorithm or set of algorithms that enable a computer to recognize patterns in a data set and interpret those patterns in actionable ways.
Supervised learning. A machine learning model that focuses its interpretation of a data set within specific parameters. A spam filter is a familiar example.
Unsupervised learning. A machine learning model that encompasses a complete data set when performing its interpretation. Data mining uses this technique.
Deep learning. A cluster of machine learning algorithms in a layered architecture, such as a neural network, that enables a high level of abstract interpretation when operating on large volumes of data. Example: Google’s AlphaGo, which recently defeated the reigning (human) champion in the board game Go.
Neural network. A programming architecture that can deal with multiple inputs and uses layers of nodes that model neurons, similar to the functioning of the nervous system.
Predictive analytics. A machine learning model that interprets patterns in data sets with the aim of suggesting future outcomes. Note: Not all predictive analytics systems use machine learning or AI-based techniques.
Oracle Management Cloud’s log analytics service, for example, uses machine learning to cluster data and identify patterns in real-time log data to spot outliers and enable administrators to troubleshoot and resolve problems quickly. Another Oracle Management Cloud service, application performance monitoring, continuously learns the behavior of each application component, such as a web server request, and detects anomalies in real time, reducing the need to manually manage threshold-based alerts across hundreds of metric streams. “Instead of just providing answers to an operator’s questions, these algorithms can continuously learn an application’s behavior and proactively provide meaningful insights when there’s a deviation from the norm,” Ganesh says.
Machine learning in Oracle Management Cloud points to another technology advance contributing to the proliferation of AI: the cloud. AI-powered applications offered as cloud services are accessible to organizations that don’t have the underlying compute and storage resources to develop them internally. And machine learning models can be offered as cloud services that enterprises can employ for their specific business or technology needs.
That’s the idea behind the Internet of Things (IoT) analytics features that are part of Oracle Internet of Things Cloud Service. “The main goal is to provide an IoT analytics platform that combines data from devices and equipment as well as other data lakes or enterprise applications,” says Bhagat Nainani, group vice president of engineering at Oracle. “Specialized analytics algorithms” analyze that integrated data and produce actionable insights that are delivered to applications downstream, he says.
For Oracle Internet of Things Cloud Service, Oracle engineers are using an open source big data toolset called Spark, which includes its own libraries of machine learning algorithms. “But it’s typically not sufficient just to have libraries,” Nainani says. “You need a data pipeline that includes the device streams, enriched with asset and customer information from multiple enterprise apps, and then you apply the analytic algorithms.”
Machine learning techniques shift the burden from the user asking the right questions to these applications finding the right context-sensitive answers.”
–Amit Ganesh, Senior Vice President, Oracle Enterprise Manager
Some machine learning algorithms developed by the Oracle Internet of Things Cloud Service team will be available out of the box, Nainani says, but a lot of them will be specially written for certain business areas and processes, such as manufacturing or supply chain.Impact and Value
The way Stephen Green describes it, his group’s job is to “find places in Oracle’s business where we think a machine learning approach can have an impact, and to implement prototypes that will show the value of these approaches. There are a lot of recent developments in algorithms and computational resources for machine learning that Oracle can be taking advantage of.” (See the “Right Place, Right Time” sidebar for more information.)
Green’s group has been particularly active in the last few years working with Oracle’s acquired companies, including Endeca, the genesis of the Oracle Big Data Discovery visualization tool, and Collective Intellect, whose social media monitoring technology is now part of Oracle Marketing Cloud.
When Oracle acquired Collective Intellect in 2012, Green realized his group had an opportunity to add significant value. Collective Intellect’s sentiment analysis system, which examined blocks of text and weighed positive and negative indicators, could be expanded and improved using more-sophisticated machine learning techniques.
Green’s group proposed a machine learning technique called classification, with which they could train the social media system “to decide whether a tweet was positive, negative, or neutral,” Green says. The technique not only improved the system’s performance, he says, but also expanded its vocabulary, his group having trained the system to understand Spanish and Chinese as well as English. The engineers then “took it and ran with it,” he adds, and the system now understands up to 10 or 11 languages.Healthy Skepticism
Despite rapid advances, Green advises a healthy dose of skepticism about the more grandiose claims made for AI—say, self-piloting jetliners. “I’ve spent a good portion of my life working on these systems,” he says. “You have to be circumspect about what you’re expecting to get out of these bigger systems over time.” There’s little doubt, though, that AI techniques such as machine learning will have a significant impact on enterprise IT architectures. Research firm International Data Corporation estimates that within only two years, about half of all applications being developed will incorporate AI.
That’s because there’s nothing artificial about the intelligence—increased effectiveness, improved processes, new competitive advantages—to be gained from employing machine learning models. But don’t take my word for it—ask Siri.
LEARN more about Oracle Internet of Things Cloud Service.
LEARN more about Oracle Management Cloud.
LEARN more about Oracle Marketing Cloud.
Photography by Fabian Fauth,Unsplash