What Is Machine Learning?

Using algorithms, computers can analyze, categorize, and make predictions

Machine Learning is an increasingly common computer technology that allows algorithms to analyze, categorize, and make predictions using large data sets. Machine Learning is less complex and less powerful than related technologies but has many uses and is employed by many large companies worldwide.

What Is Machine Learning (ML)?

Machine Learning is designed to help computers learn in ways similar to how the human brain learns. ML uses large data sets and algorithms (models) to analyze and categorize data or make predictions. The more a Machine Learning model is used, the more data it processes, the better it gets at its tasks. Models can improve on their own and can be updated by humans.

Unlike similar technologies like Deep Learning, Machine Learning doesn't use neural networks. While ML is related to developments like Artificial Intelligence), it's neither as advanced nor as powerful as those technologies.

Machine Learning has existed in various forms since the 1960s and is increasingly widely used. Around 70% of financial services companies use some form of ML by some measures.

Machine Learning Definition

Machine Learning starts with two elements: an algorithm and a data set. The algorithm instructs the ML model on what to do (analyze images, detect patterns, make predictions). The data set may or may not be classified or labeled to assist the algorithm. The algorithm then processes the data to produce an output.

The more data the algorithm processes, the more accurate it should become. Usually, the model makes the improvements based on built-in logic, but humans can also update the algorithm or make other changes to improve output quality.

This is what is meant by "learning." Humans learn basic concepts or skills and then improve through repetition and extrapolation. That's the goal of ML, too. Traditional computer programs are designed to execute a given function, but those functions are relatively limited and can only change when a programmer changes them. With ML, the model is designed to change itself based on experience with more data and tasks.

For example, an image detection algorithm might analyze pictures containing a person with red hair. The first time the model is used, its output will be less accurate than the second time, and the third time will be more accurate. This improvement happens because the model develops better techniques for distinguishing a human from a tree or a cow and distinguishing red hair from blonde hair.

Four Types of Machine Leaning

While there are sub-types within each category, the four primary types of machine learning are:

  • Supervised ML: This uses labeled, structured data and the most human intervention to find patterns the model's creator seeks. It's best used for relatively simple tasks that can be automated, with easily defined and understood rules, and where there's a lot of available data.
  • Unsupervised ML: Unlike Supervised ML, no labeled or structured data is available. Instead, the model detects patterns and draws conclusions based on the data, including things the model's creator isn't looking for. It's used to detect patterns and for classification (such as grouping customers based on behavior) and then to take actions based on those patterns.
  • Semi-Supervised ML: Combines the two types above by initially training the model with labeled data and then having the model process unlabeled data. Semi-supervised ML is helpful when there's not enough labeled data or generating that data is impractical.
  • Reinforcement: This type of ML is based on rewards or positive feedback and is best used for systems where you can easily define right/wrong answers and where there's a best action in a given. ML models that play games like chess are often reinforcement models.

Common Uses for Machine Learning

Machine Learning applications that many people have experienced include these common uses:

  • Recommendation Algorithms: ML models' pattern detection and categorization features are at the heart of algorithms recommending content or products.
  • Speech Recognition: ML is used for text-to-speech software and natural-language processing applications.
  • Chatbots: Chatbots used for sales and customer service, especially those that are relatively simple decision trees, are often based on ML.
  • Computer Vision: Giving computers the ability to "see" and understand images—for things as complex as autonomous vehicles or as simple as analyzing photos—relies on Machine Learning.
  • Fraud Detection and Anti-Spam: Pattern detection helps financial services companies flag potentially fraudulent transactions and allows your email account to remove spam from your inbox.

Areas of Concern for Machine Learning

While ML is powerful and widely used, it has also come under criticism for issues like:

  • Privacy: Because ML models need large volumes of data, there's the chance that ML could process sensitive personal data. In some cases, the use of that data may not have been authorized or fully understood by individuals. As a result, ML could use or disclose sensitive information.
  • Lack of Transparency: Because of their ability to learn, it can be very hard—sometimes impossible—to understand each step that led a model to a conclusion or recommendation. That limitation makes it very difficult to regulate to ensure they're not violating laws (such as fair housing or lending laws).
  • Bias and Discrimination: Bias (conscious or unconscious) on the part of model creators or in the data used to train the model will influence how the model learns and its outputs. So, unless special attention is paid to correcting bias, ML models can unintentionally reinforce bias. A good example is how facial recognition systems are more accurate with some skin tones than others based on the data used to train them and what race the models' creators belong to. This can lead to significant problems in law enforcement, for instance.
  • What is overfitting in machine learning?

    Overfitting is an error with the output of a machine learning program. It generally means that the output too closely resembles (fits) the training data, which suggests that the algorithm isn't doing the inference or extrapolation it should be and therefore isn't producing usable results.

  • What is cross-validation in machine learning?

    Cross-validation is a method of testing a machine-learning model; developers usually use it to combat overfitting. One version of cross-validation involves splitting the original data set into smaller chunks. Some are held back, while the others run through the model. You then compare the "control" groups with the "test" groups to how the algorithm performs.

Was this page helpful?