078-2034021 algemeen@3bplus.nl

In a previous article we gave a definition of machine learning. Subsequently, we will now discuss machine learning methods: supervised learning, unsupervised learning, semi-supervised learning and reinforcement learning.

Methods of machine learning – different methods of learning

The most commonly used methods of machine learning are supervised learning, unsupervised learning, semi-supervised learning and reinforcement learning. Although all of these methods have the same goal – to arrive at insights, patterns, and relationships that can be used to make decisions – they use different approaches.

Methods of machine learning – supervised learning

With machine learning, a machine is trained to see connections. This is often done through supervised learning, or controlled learning. With supervised learning, the computer receives exemplary input, whereby the desired output is also presented. The goal is to learn a general rule that translates the given input into this desired output. The system learns to see connections between the input and the output. If the machine learning process is followed properly, the system makes fewer and fewer errors and can ultimately produce the correct output based on new input.

Supervised learning algorithms are trained using labeled examples: input for which the desired output is known. The algorithm learns by comparing with the correct output, so that the model is adjusted where necessary. This model is then used to predict the output (the value of the label) for non-labeled input. In other words: supervised algorithms can apply what has been learned in the past to new data. Supervised learning is therefore often used in applications where historical data can predict future events, such as fraudulent credit card transactions.

Text continues after image.

Methods of machine learning – unsupervised learning

When applying unsupervised learning to machine learning, no labels are given to the input of the learning algorithm. In this uncontrolled learning, the guidance of entering examples with a desired output is not offered. With unsupervised learning, the algorithm will eventually discover a structure in the input itself. Due to the lack of labels, the learning algorithm is dependent on itself to find structure in the input and to come up with the correct output. In other words: the algorithm must find out what is being displayed without labels (the correct output). During this process, the computer itself will divide the input into categories. Elements are then placed with data that closely resemble each other.

Unsupervised learning is therefore used with data that does not have historical labels, such as with transaction data. With this data it can, for example, identify customer segments with similar characteristics, which is useful for marketing campaigns. These algorithms are also used to recommend items.

Methods of machine learning – semi-supervised learning

Semi-supervised learning in machine learning is basically used for the same applications as supervised learning. Semi-supervised learning, however, uses both labeled and non-labeled data for training – usually a small amount of labeled data with a large number of non-labeled data. Thi methods is chosen because unlabeled data is less expensive and less difficult to acquire. Semi-supervised learning is therefore useful if the costs associated with labeling labeling are too high to enable a fully labeled training process.

Methods of machine learning – reinforcement learning

Reinforcement learning in machine learning has three main components: the agent (the learner or the decision maker), the environment (everything the agent interacts with) and actions (what the agent can do). The agent navigates in a dynamic environment in which it must perform a certain purpose, such as playing a game. The goal is to let the agent choose actions that maximize the expected reward over a certain time. The agent achieves the goal much quicker by following a good policy. In other words: the goal with reinforcement learning is to learn the best policy. The agent is thus given feedback (rewards and penalties), while navigating within the problem area. With reinforcement learning, the algorithm discovers through trial and error which actions yield the greatest rewards. Reinforcement learning is often used for robotics, gaming and navigation.

You can find more English articles in our blog.

Other articles you might enjoy:

What is machine learning: An introduction
Machine learning, neural networks and deep learning explained
How do smart devices work: sensors, IoT, Big Data and AI

Share This