The term machine learning (ML) refers to the use of advanced mathematical models—typically referred to as algorithms—to process large volumes of data and gain insight without direct human instruction or involvement.
ML is a subset of artificial intelligence (AI). It is built on artificial neural networks (ANNs) or simulated neural networks (SNNs)—essentially node layers that interact and interconnect. It includes a specialized type of machine learning called deep learning (DL).
Machine learning mimics the way humans learn. It spots patterns and then uses the data to make predictions about future behavior, actions and events. In addition, ML constantly uses new data to adapt and change its actions. This ability to learn from experience separates it from more static tools such as business intelligence (BI) and conventional data analytics.
Organizations across numerous fields are turning to ML to address complex business challenges. The technology is particularly valuable in areas such as marketing and sales, financial services, healthcare, retail, energy, transportation and government planning. High profile examples of organizations using machine learning include Netflix, Uber, Google, Facebook and Amazon. The technology handles tasks as diverse as pricing, delivery times, search results and product recommendations.
Depending on the use case, ML requires specific training methods in order to function effectively—and deliver value. These approaches include supervised and unsupervised learning, which means the system learns with humans overseeing it or on its own.
Today, machine learning is used for tasks as varied as speech recognition, image detection and machine vision, predicting customer behavior, spotting fraud and cybersecurity threats and overseeing machine maintenance.
Also see: Best Machine Learning Platforms
How are Machine Learning Methods Used?
Businesses, governments, educational institutions and many other entities rely on ML to deliver guidance and make key decisions. In many cases, ML system are incorporated into broader automation and AI frameworks. This might include a smart transportation system that automatically adapt to conditions, such as weather, traffic and other events.
Another example is sentiment analysis, which plugs in different data—historical buying patterns, current data about raw materials and pricing, weather conditions, social media trends and more—to generate a model that predicts future pricing and buying, even under specific conditions.
In addition, ML is now used to develop and improve performance many ways. ML will:
- Enhance smart speakers and personal assistants on smartphones.
- Detect unsafe behavior in factories.
- Allow airline passengers to board planes and go through passport control using biometrics.
- Develop robots, digital twins and other business tools that continually learn and improve as data is added.
Consulting firm Gartner reports that top use cases revolve around five core areas: knowledge management, virtual assistants, autonomous vehicles, the digital workplace, and crowdsourced data. Adoption is accelerating rapidly as digital transformation becomes a growing focus.
Worldwide artificial intelligence (AI) software revenue, including machine learning, is forecast to total $62.5 billion by the end of 2022, an increase of 21.3% from 2021, it noted.
Also see: Top AI Software
A Brief History of Machine Learning
The idea that machines could learn and adapt their algorithms was introduced by logician Walter Pitts and neuroscientist Warren McCulloch, who published a research paper outlining the concept in 1943.
In 1950, computer scientist Alan Turing introduced the Turing Test, also referred to as the “imitation game,” a framework that gauges a machine’s ability to display intelligent behavior indistinguishable from humans.
The words machine learning were coined by IBM data scientist Arthur L. Samuel in 1959. In an academic paper, he promoted the idea that a computer could learn to play checkers and compete with humans. Samuel developed an algorithm that learned to play the game without explicit programming. In 1962, a checkers master, Robert Nealey, played against an IBM 7094 computer and lost.
Over the last 60 years, the ML frameworks have grown and expanded. Far greater computational power along with new and different types of statistical methods, or algorithms, have led to radical advances in the field.
As ML has evolved, explanation-based learning has been replaced by neural nets and deep learning methods that are less explainable. In 2009, the emergence of convolutional neural nets (CNN) revolutionized the field. In 2011, IBM’s Watson, a CNN, beat human competitors in the television show Jeopardy.
These CNNs process multiple layers of data—much like the human brain. CNNs handle mathematical learning and computational processes behind the scenes on their own and allow filtering and tuning in real time. Today, CNNs are used for advanced tasks such as facial recognition and live language translation. Companies such as Netflix, Google, Apple and many others used CNNs and their cousin, Generative Adversarial Networks (GAN) to handle increasingly complex ML and AI tasks.
How Do Machine Learning Systems Work?
Four primary types of ML methods exist:
- Supervised learning, which requires a person to identity the desirable signals and outputs through labeling or classification.
- Unsupervised learning, which allows the system to operate independent of humans and find valuable output using unlabeled data.
- Semi-supervised learning, which combines the two methods above.
- Reinforcement learning, which incorporates a computer program that interacts with a dynamic environment to achieve specific goals and outcomes.
Machine learning frameworks use software languages such as TensorFlow and PyTorch to deliver a usable model. According to UC Berkeley researchers, an ML model involves three distinct components:
- A Decision Process. A system ingests data and uses a machine learning algorithm to classify and predict events.
- An Error Function. This built-in capability allows the model to evaluate the accuracy and quality of predictions.
- A Model Optimization Process. This feature allows ML models to adapt, based on finding data that’s a better fit with the training set. As the model changes, the system continues to evaluate and refine itself to achieve accuracy goals.
What Types of Machine Learning Frameworks are Used?
Machine learning revolves around several core algorithmic frameworks to achieve results and produce models that are useful. These include:
These systems are comprised of artificial intelligence algorithms that are designed to simulate the way the human brain thinks. They use training data to spot patterns, and they typically learn rapidly using thousands or even millions of processing notes. They’re ideal for recognizing patterns and they are widely used for speech recognition, natural language processing, image recognition, consumer behavior and financial predictions.
The technique identifies relationships between independent input variables and at least one target variable. It is valuable for predicting numerical values, such as prices for airline flights or real estate values, usually over a period of weeks or months. It can display predicted price or value increases and decreases across a complex data set.
This method typically uses a binary classification model (such as “yes/no”) to tag or categorize whether an event is likely to occur. It sorts through a dataset to find weights and biases that can be built into or excluded from the model. For instance, a common use for this technology is identifying spam in email and blacklisting unwanted software code or malware.
This ML tool uses unsupervised learning to spot patterns and relationships that humans may overlook. An example of clustering is how a supplier performs for the same product at different facilities. This approach might be used in healthcare, for instance, to understand how different lifestyle conditions impact health and longevity. It can also be used for trend detection at websites and in social media, such as what text, images and video to display.
The supervised learning approach builds a data structure with nodes that test an idea or concept against a set of input data. A Decision Tree delivers numerical values but also performs some classification functions. It helps users visually understand data. Unlike other forms of ML, it makes it possible to review and audit results. In the business world, decision trees are often used to develop insights and predictions about downsizing or expanding, changing a pricing model or succession planning.
A Random Forest model incorporates multiple decision tree models simultaneously. Combining decision trees makes it possible to classify categorical variables or the regression of continuous variables—forming what’s called an ensemble. This makes it possible to use different trees to produce specific predictions but then combine the predictions into a single ensemble or overall model. A random forest algorithm ML tool might be used for a recommendation system, for example.
What is Deep Learning?
Neural nets serve as the foundation for deep learning models—which in turn feed many of today’s AI systems. Deep learning systems rely on interconnected layers of machine learning algorithms—typically through graphic processing units (GPUs)—to develop and continuously evolve a model. While a more basic neural net incorporates one or two hidden layers, a DL model may include dozens, hundreds or even thousands of layers.
For instance, if a deep learning system is trained on birds, it learns how to distinguish eagles from hawks, crows from ravens, and chickadees from hummingbirds. Deep learning frameworks are fast and tend to deliver sophisticated models that improve automation systems and tools such as Apple’s Siri or Amazon’s Alexa, TV remotes, credit card fraud detection, captions for YouTube videos and autonomous vehicle behavior.
Also see: The Future of Artificial Intelligence
How do ML and AI Differ?
Unfortunately, ML and AI are often used synonymously. Attempting to distinguish between the two fields can be difficult, partly because they overlap. However, a starting point is recognizing that ML is always a subset of AI.
In a broad sense, artificial intelligence attempts to simulate human thinking and behavior. Machine learning specifically relates to systems that learn about conditions through data without a human interface and then apply the data to decision-making and other events, such as automation.
How is ML Evolving in the Enterprise?
As Gartner noted, ML adoption is growing rapidly. The technology is increasingly incorporated into enterprise software applications, smartphone apps, and it is available as a discreet service through cloud platforms from the likes of AWS, Google, Microsoft and others.
Tools are becoming easier to use—in many cases they’re now available in low-code and no-code platforms—thus expanding their availability to line of business users as well as data scientists. Yet, no matter how sophisticated ML platforms become, they require human oversight, including a strategic focus on how to use them effectively. If an ML tool is poorly constructed or an organization feeds it with low quality data, the results can be useless and even damaging.
Also see: The History of Artificial Intelligence
What Ethical and Legal Concerns Exist?
Bias is a growing concern with ML systems. Depending on underlying data used for training, they can generate discriminatory and bias results.
In recent years, these systems have been associated with hiring bias and overall gender and racial bias—including among law enforcement and government agencies. The data can also be misused when it is fed into broader AI systems that generate false news and touch on areas such as surveillance, robotics, marketing and advertising.
Since there’s virtually no regulation of AI, organizations should have a team overseeing ML and AI ethics policies and data privacy standards internally. It’s also important to tune into broader Ethical AI trends in the business world.
What is the Future of Machine Learning?
Rapid advancement of ML technology ensures that it will play an increasingly prominent role in defining business in the years to come. It will impact agriculture, finance, manufacturing, transportation, marketing, customer support, cybersecurity and many other areas. Machine learning will also help drive corporate Environmental, Social, and Governance (ESG) programs and sustainability initiatives. These initiatives will affect sourcing, supply chains and Scope 3 emissions that extend back to raw materials and component providers.
Machine learning systems are becoming easier to use and manage. As a result, they are extending deeper into organizations and moving beyond the realm of data scientists. As organizations looks to trim costs, boost productivity, oversee ESG programs, build smart factories, better manage supply chains and fuel innovation at scale, ML emerges as an essential tool.
Savvy business and IT leaders now look for ways to adopt and expand the use of machine learning while exploring test cases that could unlock transformative gains in the future.
Also see: What is Artificial Intelligence