You probably heard more and more about machine learning, a subset of artificial intelligence. But what exactly can be done with it? Technology encompasses methods, and each has a set of potential use cases.
Companies would do well to examine them before moving forward with plans to invest in the machine learning tools and the infrastructure.
This Supervised learning is ideal if you know what you want a machine to learn. You can expose it to a vast set of training data, examine the result, and modify the parameters until you get the results you expect. After that, you can see what the machine has learned by predicting the results of a validation data set that you have not seen before.
The supervised learning tasks include classification and prediction or regression. Supervised learning methods are often used for applications like determining the financial risk of people and organizations, supported past information on financial performance. They can also provide a good idea of how customers will act or what their preferences are based on previous behavior patterns.
For example, the Lending Tree online loan market is using an automated DataRobot machine learning platform to customize experiences for its customers and predict their intention based on what they have done in the past, says Akshay Tandon, vice president and chief of strategy and analytics.
By predicting the intention of the client – primarily through the qualification of potential clients – Lending Tree can classify people who are looking for a rate, compared to those who are really looking for a loan and are ready to apply for one. Using supervised learning techniques, he constructed a classification model to define the probability of a lead closing.
Unsupervised learning allows a machine to explore a set of data and identify hidden patterns that link different variables. This method can be used to group data into groups only based on their statistical properties.
A useful application of the unsupervised learning is the clustering algorithm used to make links to probabilistic records, a technique that extracts connections between data elements and builds them to identify people and organizations and their connections in the physical or virtual world.
This is especially useful for companies that need, for example, to integrate data from disparate sources and/ or in different business units to build a consistent and complete view of their customers, says Flavio Villanustre, vice president of technology at LexisNexis Risk Solutions, A company that uses analytics to help customers predict and manage risks.
Unsupervised learning could be used for sentiment analysis, which identifies people’s emotional state based on their social media posts, emails, or other written comments, notes Sally Epstein, a machine learning engineering specialist at Cambridge consultancy. Consultants the firm has gotten an increasing number of companies in financial services that use unsupervised learning to obtain information on customer satisfaction.
This is a hybrid of supervised and unsupervised learning. By labeling a little portion of the data, a trainer can give the machine clues about how to group the rest of the data set.
Semi-supervised learning can be used to detect identity fraud, among other uses. Fortunately, fraud is not as frequent as a non-fraudulent activity, Villanustre notes, and as such, fraudulent activity can be considered an “anomaly” in the universe of legitimate activity. Even so, there is a fraud, and machine learning methods for detecting semi-supervised anomalies can be used to model solutions to these types of problems. This type of learning is implemented to identify fraud in online transactions.
Semi-supervised learning can also be used when there is a mix of labeled and unlabeled data, which is often seen in large business environments, Epstein notes. Amazon has been able to improve the natural language understanding of its Alexa offering by training artificial intelligence algorithms in a combination of labeled and unlabeled data, says the executive. This has helped increase the accuracy of Alexa’s responses, he says.
Reinforcement learning allows the machine to interact with its surroundings (for example, pushing damaged products from a conveyor to a container) and provides you with a reward when you do what you want. By automating such a calculation of the reward, you can let the machine learn on its own time.
A use case for reinforcement learning is the classification of clothing and other items in a retail establishment.
Some clothing retailers are testing new sorts of technology, like robotics, to assist classify items like clothing, footwear, and accessories, says David Schatsky, an analyst at Deloitte, which focuses on emerging technology and business trends.
Robots make use of reinforcement learning (as well as deep learning) to determine how much pressure they should use when grabbing objects and the best way to grab these items in the inventory, Schatsky notes.
A variation of reinforcement learning is the deep reinforcement learning, which is very suitable for autonomous decision making, where supervised learning or unsupervised learning techniques alone cannot do the job.
Deep learning performs types of learning, such as unsupervised or reinforced. In general terms, deep learning mimics some aspects of how people learn, mainly through the use of neural networks to identify the characteristics of the data set in more and more details.
Deep learning, within the form of deep neural networks (DNN), has been used to accelerate high-content screening for drug discovery, says Schatsky. It is about applying DNN acceleration techniques to process multiple images in significantly less time while extracting a higher perception of the characteristics of the image that the model finally learns.
This machine learning method also allows many companies to fight fraud, improving detection rates by using automation to detect irregularities. Deep learning can be used in the automotive industry. A company has developed a system based on neural networks that allow early detection of problems with cars, says Schatsky. This system can recognize noise and vibration and use any deviation from the norm to interpret the nature of the fault. It can become part of predictive maintenance since it determines the vibrations of any mobile part of the car and you may notice even minor changes in its performance.