Models of Machine Learning

Supervised learning

Supervised learning is the simplest of the learning models to understand. Learning in the supervised model entails creating a function that can be trained by using a training data set, then applied to unseen data to meet some predictive performance. The goal is to build the function so that it generalizes well over data it has never seen.

Neural networks

A neural network processes an input vector to a resulting output vector through a model inspired by neurons and their connectivity in the brain. The model consists of layers of neurons interconnected through weights that alter the importance of certain inputs over others. Each neuron includes an activation function that determines the output of the neuron (as a function of its input vector multiplied by its weight vector). The output is computed by applying the input vector to the input layer of the network, then computing the outputs of each neuron through the network (in a feed-forward fashion).

Decision trees

A decision tree is a supervised learning method for classification. Algorithms of this variety create trees that predict the result of an input vector based on decision rules inferred from the features present in the data. Decision trees are useful because they’re easy to visualize so you can understand the factors that lead to a result.

Unsupervised learning

Unsupervised learning is also a relatively simple learning model, but as the name suggests, it lacks a critic and has no way to measure its performance. The goal is to build a mapping function that categorizes the data into classes based on features hidden within the data.

K-means clustering

k-means clustering is a simple and popular clustering algorithm that originated in signal processing. The goal of the algorithm is to partition examples from a data set into k clusters. Each example is a numerical vector that allows the distance between vectors to be calculated as a Euclidean distance.

Adaptive resonance theory

Adaptive resonance theory (ART) is a family of algorithms that provide pattern recognition and prediction capabilities. You can divide ART along unsupervised and supervised models, but I focus here on the unsupervised side. ART is a self-organizing neural network architecture. The approach allows learning new mappings while maintaining existing knowledge.

Reinforcement learning

Reinforcement learning is an interesting learning model, with the ability not just to learn how to map an input to an output but to map a series of inputs to outputs with dependencies (Markov decision processes, for example). Reinforcement learning exists in the context of states in an environment and the actions possible at a given state. During the learning process, the algorithm randomly explores the state–action pairs within some environment (to build a state–action pair table), then in practice of the learned information exploits the state–action pair rewards to choose the best action for a given state that lead to some goal state. You can learn more about reinforcement learning in “Train a software agent to behave rationally with reinforcement learning.”


Q-learning is one approach to reinforcement learning that incorporates Q values for each state–action pair that indicate the reward to following a given state path. The general algorithm for Q-learning is to learn rewards in an environment in stages. Each state encompasses taking actions for states until a goal state is reached. During learning, actions selected are done so probabilistically (as a function of the Q values), which allows exploration of the state-action space. When the goal state is reached, the process begins again, starting from some initial position.

Going further

Machine-learning benefits from a diverse set of algorithms that suit different needs. Supervised learning algorithms learn a mapping function for a data set with an existing classification, where unsupervised learning algorithms can categorize an unlabeled data set based on some hidden features in the data. Finally, reinforcement learning can learn policies for decision-making in an uncertain environment through iterative exploration of that environment.



Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Aditya B

Aditya B


Passionate author, strategic investor, financial advisor