What Does Machine Learning (ML) Mean?
Machine learning (ML) is the sub-category of artificial intelligence (AI) that builds algorithmic models to identify patterns and relationships in data. In this context, the word machine is a synonym for computer program and the word learning describes how ML algorithms become more accurate as they receive additional data.
The concept of machine learning is not new, but its practical application in business was not financially feasible until the advent of the internet and recent advances in big data analytics and cloud computing. That’s because training an ML algorithm to find patterns in data requires a lot of compute resources and access to big data.
The terms artificial intelligence and machine learning are sometimes used as synonyms because until recently, most AI initiatives have been narrow and most ML models were built to perform a single task, used supervised learning and required large, labeled data sets for training. Today, robotic process automation (RPA) can be used to automate the data pre-processing process and make training a machine learning algorithm much faster.
Techopedia Explains Machine Learning (ML)
High-quality machine learning models require high-quality training data and access to large data sets in order to extract features most relevant to specified business goals and reveal meaningful associations.
Machine Learning Models
A machine learning model is simply the output of an ML algorithm that has been run on data. The steps involved in building a machine learning model include the following:
- Gather training data.
- Prepare data for training.
- Decide which learning algorithm to use.
- Train the learning algorithm.
- Evaluate the learning algorithm’s outputs.
- If necessary, adjust the variables (hyperparameters) that govern the training process in order to improve output.
In a typical ML setting, supervised machine learning algorithms require a dataset comprised of examples where each example consists of an input and output. In such a setting, a typical objective of training a ML algorithm is to update the parameters of a predictive model to ensure the model’s decision trees consistently produces desired outcomes. This is where entropy comes in.
Entropy is a mathematical formula used to quantify the disorder and randomness in a closed system. In machine learning projects, an important goal is to make sure entropy remains as low as possible because this measure will determine how the model’s decision trees will choose to split data.
Training Machine Learning
There are three main types of algorithms used to train machine learning models: supervised learning, unsupervised learning and reinforcement learning.
- Supervised learning – the algorithm is given labeled training data (input) and shown the correct answer (output). This type of learning algorithm uses outcomes from historical data sets to predict output values for new, incoming data.
- Unsupervised learning – the algorithm is given training data that is not labeled. Instead of being asked to predict the correct output, this type of learning algorithm uses the training data to detect patterns that can then be applied to other groups of data that exhibit similar behavior. In some situations, it may be necessary to use a small amount of labeled data with a larger amount of unlabeled data during training. This type of training is often referred to as semi-supervised machine learning.
- Reinforcement learning – instead of being given training data, the algorithm is given a reward signal and looks for patterns in data that will give the reward. This type of learning algorithm’s input is often derived from the learning algorithm’s interaction with a physical or digital environment.
What Causes Bias in Machine Learning?
There is a growing desire by the general public for artificial intelligence – and machine learning algorithms in particular — to be transparent and explainable, but algorithmic transparency for machine learning can be more complicated than just sharing which algorithm was used to make a particular prediction.
Many people who are new to ML are surprised to discover that it’s not the mathematical algorithms that are secret; in fact, most of the popular ML algorithms in use today are freely available. It’s the training data that has proprietary value, not the algorithm used.
Unfortunately, because the data used to train a learning algorithm is selected by a human being, it can inadvertently introduce bias to the ML model that’s being built. The iterative nature of learning algorithms can also make it difficult for ML engineers to go back and trace the logic behind a particular prediction.
When it is possible for a data scientist or ML engineer to explain how a specific prediction was made, an ML model may be referred to as explainable AI. When it is not possible to reveal how a specific prediction was made — either because the math becomes too complicated or the training data is proprietary — the ML model may be referred to as black box AI.
Machine learning projects are usually overseen by data scientists and machine learning engineers. The data scientist’s job typically involves creating an hypothesis and writing code that will hopefully prove the hypothesis to be true. An ML engineer’s job focuses on machine learning operations (MLOps).
Machine learning operations is an approach to managing the entire lifecycle of a machine learning model — including its training, tuning, everyday use in a production environment and eventual retirement. This is why ML engineers need to have a working knowledge of data modeling, feature engineering and programming — In addition to having a strong background in mathematics and statistics.
Ideally, data scientists and ML engineers in the same organization will collaborate when deciding which type of learning algorithm will work best to solve a particular business problem, but in some industries the ML engineer’s job is limited to deciding what data should be used for training and how machine learning model outcomes will be validated.
|Free Download: AI in the Insurance Industry: 26 Real-World Use Cases|