While we aren’t yet dealing with artificial intelligence like Hal from Stanley Kubrick’s 2001: A Space Odyssey or Samantha from Spike Jonze’s Her, AI technologies are increasingly impacting our everyday lives. This is especially the case for machine learning: from Google Assistant, to UberEATS’s estimated time-of-delivery model, to Facebook’s pivot to feature AI-generated recommendations on your newsfeed, machine learning is making our world more connected, more efficient, and more personalized.
It’s no surprise, then, that demand for machine learning solutions is growing at a stunning rate: by 2028, Grand View Research projects the global AI market to expand by over 40%, with revenues forecasted to top $900 trillion worldwide. Demand for artificial intelligence and machine learning degrees and bootcamps has been growing to match. At the 18 universities worldwide surveyed by Stanford’s AI Index, the number of courses offering undergrads instruction in developing practical AI and ML models has grown by 102.9% over four academic years from 2016 to 2020. Enrollment at these schools grew considerably with respect to introductory courses in artificial intelligence and machine learning, increasing by 60% over the same time-frame.
While technologies employed at big companies like Google, Meta, Uber, and Spotify are surely top-of-mind for students just getting into artificial intelligence and machine learning, the remarkable growth in the field suggests that most graduates in the next ten years will be tasked with implementing AI and ML solutions at an ever-larger number of companies across an ever-wider variety of industries.
In this article, we preview some of the machine learning methods driving this growth before going into greater detail about the ways machine learning is making a difference in finance, healthcare, and manufacturing. At the end, we’ll suggest some ways you can get involved in this groundbreaking work.
What are the most popular machine learning methods?
Supervised learning refers to the use of labeled data sets — or “training sets” — to train a machine learning algorithm to give the correct output when fed an input. In essence, these algorithms build an ML model that learns by example. A common use of supervised machine learning is for text recognition in natural language processing. As an algorithm is fed many images of letters and told what letters the images represent, it learns to identify these letters without prompting. Supervised learning algorithms are also employed for various types of speech recognition.
Within supervised machine learning, IBM identifies two common techniques, regression and classification. Both techniques are common for data mining, where big data sets are probed for interesting relationships, patterns, or outliers. For data analysis, both regression and classification use decision trees: rules-based steps that an algorithm will move through to decide on a particular data point.
A decision tree for settlement classification from an article in the International Journal of Geo-Information.
But while regression and classification share a work process, there are fundamental differences in what the techniques seek to accomplish:
Though there are many types of regression used in machine learning, in essence, this ML technique boils down to using independent, known variables to predict the outcome of a dependent variable. While there are countless applications for regression, one area where regression is proving particularly useful is in climate change research. Recently, researchers in India used various types of regression to analyze data for various climate change parameters and develop models to predict these parameters in the future.
Whereas regression is interested in establishing relationships between variables, classification focuses on sorting input data into various output groups. A common application is in spam filters. After being fed training data and instructed on which examples are “spam” or “not spam,” an algorithm learns to discern for itself which emails should land in your inbox and which should be siloed in your spam folder.
Unsupervised learning refers to the use of unlabeled data sets to train a machine learning algorithm. Rather than learning by example, unsupervised ML algorithms are written to be able to make sense of the data themselves: discovering patterns and relationships, forming clusters, and cutting through the data to get to the most important data points. Recently, a group of researchers detailed in Nature how they used unsupervised learning to distinguish groups of Alzheimer’s and Parkinson’s Disease patients based on their molecular make-up rather than on how the diseases were presenting.
IBM divides unsupervised learning into three main techniques: clustering, association, and dimensionality reduction:
In clustering, points from an unlabeled data set are sorted according to their similarities and differences. It is commonly used for tasks like market segmentation, online recommendation, search results, and medical imaging analysis.
Association, or “association rule learning,” asks machine learning algorithms to search big data sets or databases for interesting or useful relationships. These relationships are discovered by inputting a set of “if/then” rules into the algorithm, as well as a numerical criterion for how interesting, useful, or essential a found relationship might be. Association rule learning is commonly used for affinity or “market basket” analysis that looks for patterns in the kinds of products consumers buy. This kind of analysis is useful for everything from “Customers also Buy” product recommendation features to algorithm-generated music and video playlists.
In the era of big data, with massive amounts of data being taken in at every moment, sometimes a data set will need to be simplified before undergoing further analysis. A solution comes in the unsupervised learning technique of dimensionality reduction, in which a machine learning algorithm will analyze a large dataset and remove unnecessary or irrelevant parameters, or “dimensions.” Perhaps the most prominent example of a dimensionality reduction algorithm is the autoencoder, a type of deep learning neural network that learns to represent only the necessary elements of a dataset, without any so-called noise. A common use of autoencoders is in image or audio compression. For a deep dive on neural networks and autoencoders, check out our deep learning article.
Reinforcement learning differs from supervised and unsupervised learning in that it is explicitly goal-oriented, with machine learning algorithms written to behave in ways that will maximize a numerical reward as it engages with a complex environment. Reinforcement learning has application throughout industry, including in teaching cars how to drive themselves and in optimizing robots for automated industrial tasks.
Reinforcement learning usually occurs through something called a Markov decision process:
Markov Decision Process
A Markov decision process is used in machine learning models to accommodate both an uncontrollable environment and an AI’s sequential choices within that environment. As an artificial intelligence agent continues to interact with an environment, it receives constant feedback in the form of numerical rewards — numbers that tell the AI that it’s on the right track. Reinforcement learning algorithms are written in such a way that making a series of decisions will both maximize the numerical reward and achieve the goal desired by the programmer. One useful application of Markov decision processes is in dam and reservoir management, where these processes are used to maximize power output of a hydroelectric generator while managing water-levels.
A visualization of a Markov decision chain from a 2018 article in Nature.
How is machine learning driving growth today?
You’ve just seen some examples of how these various machine learning techniques can be employed in certain situations, but what is the bigger picture for machine learning applications? How is machine learning making a difference not just for discrete tasks, but across entire industries?
Machine Learning in Finance
In the coming years, the Organisation for Co-Operation and Development (OECD) forecasts that artificial intelligence will boost the efficiency of financial firms by reducing costs and increasing productivity, as well as allowing for the launch of improved products and services. Particular areas of the financial services sector that the OECD suggests will benefit from artificial intelligence and machine learning include:
There are a variety of machine learning applications that support asset management. ML models trained on big data, for example, can offer recommendations for asset allocation, while natural language generation, a relative of natural language processing, can provide analysis and reporting more easily understood by human clients. With its ability to quickly play out countless potential scenarios, machine learning can also present opportunities for improved risk management.
Predictive analytics powered by machine learning can allow for more reliable creditworthiness assessments.
Artificial intelligence can allow for more sophisticated trade automation that improves on recent algorithmic high-frequency trading models. For its part, machine learning can offer more accurate sentiment analysis and improved predictive capabilities.
While artificial intelligence and machine learning technologies have not yet been merged with blockchain technologies at scale, the OECD believes a merger promises to improve security, efficiency, and data management.
You can dive deeper into these kinds of possibilities by checking out our article on machine learning in finance.
Machine Learning in Healthcare
In an article in Future Healthcare Journal, Thomas Davenport and Ravi Kalakota identify “diagnosis and treatment recommendations, patient engagement and adherence, and administrative activities” as the primary areas in healthcare where artificial intelligence is making a difference. For machine learning in particular, they note several exciting applications:
The most common application of “traditional” machine learning, according to Davenport and Kalakota, is in precision medicine, the practice of predicting which treatments will work for a patient given their background.
Deep learning, the more advanced subdiscipline of machine learning that leverages artificial neural networks, frequently teams up with computer vision technologies in radiomics, in which medical images are broken down into quantitative data that can be analyzed to support decision-making.
Patient engagement and adherence
Doctors can only do so much, but machine learning is increasingly helping patients help themselves. Using machine learning algorithms to manage “messaging alerts and relevant, targeted content that provoke actions at moments that matter,” healthcare professionals are increasingly finding ways to expand their influence and improve patient compliance.
While perhaps not as groundbreaking as applications in other areas, machine learning can also provide important improvements in healthcare administration. Natural language processing can drive efficiency in patient-provider communications, while machine learning-driven data mining can streamline the insurance claim process.
In our article on machine learning in healthcare, we’ll give you specific examples of how these kinds of machine learning applications are helping patients worldwide.
Machine Learning in Manufacturing
According to Tulip, an operations platform, machine learning is currently leveraging the vast amount of data created in goods production to provide crucial insights into how to streamline manufacturing processes. Particular areas of improvement include:
Machine learning models are able to find patterns in data that signal potential machine failure. When these machines are fixed before they break, manufacturers avoid costly downtime.
Energy utilization and prediction
By analyzing energy data, machine learning models are able to find areas of inefficient energy consumption and predict future energy needs, both of which reduce costs.
Supply chain management
According to McKinsey, consumer packaged goods (CPG) companies are now able to automate their supply chains end-to-end, using machine learning models to manage inventory, predict future demand, and increase production efficiency.
If you’re interested in learning more about how machine learning models are getting more products to more consumers and businesses for less money, check out our article on machine learning in manufacturing.
How can you get involved in machine learning?
As you can see, machine learning is much, much more than digital assistants and personalized playlists: it’s increasingly driving almost every important industry in the world. If you want to play a role in machine learning’s next chapter, check out our articles on the different jobs in artificial intelligence and machine learning to see what’s involved in being a machine learning engineer, data scientist, or AI software developer.
If you’re ready to take the next step down one of these career paths, the most common route is to get a bachelor’s degree in computer science, artificial intelligence, or machine learning. This will put you well on your way to becoming a machine learning engineer or data scientist, and you can check out our recommendations for some great bachelor’s programs to see if one might be a good fit. You can also try a data science or machine learning bootcamp if you’re looking for a shorter-term option.
Again, we have some great recommendations to help you make the best decision for you. If you already have a bachelor’s degree in a STEM field, you might consider applying to a master’s program in computer science, AI, or machine learning — and you guessed it, we have some programs we think you’ll love.