Machine Learning

    What is machine learning and how is it used? You can find out more here.

    Definition

    Machine learning is a central subfield of AI that deals with letting computer algorithms learn from data instead of explicitly programming them. A machine learning model is trained with sample data (and often desired outputs) and adapts until it recognizes patterns in the data. It can then generalize these learned patterns to make meaningful predictions or decisions based on new data.

    In short: the system learns from experience. Machine learning thus enables image recognition, language translation or predictions of trends, for example, without a human having to specify all the rules.

    Learning types at a glance

    There are different types of machine learning:

    • Supervised learning: During training, the model receives inputs and the corresponding correct outputs (labels). It uses these examples to learn a general function. Example: Images of cats and dogs (input) with the label "cat" or "dog" – the model learns to classify new images correctly.
    • Unsupervised learning: Only input data is given, no labels. The model should discover structures in the data itself, e.g. clusters or correlations. Example: A large collection of customer information – the model groups the customers according to similarities, without predefined categories.

    • Reinforcement learning: An agent learns through interaction with the environment and rewards. There is no static training data set here; the agent gathers experience in a sequence of states.

    • Self-Supervised Learning: A newer important class where labels are quasi obtained from the data itself (e.g. filling gaps as an exercise). This is how LLMs are trained.

    Applications

    Machine learning is omnipresent today: recommendation systems from online stores or streaming services learn to make suitable suggestions based on your behavior. Spam filters learn to recognize unwanted emails. Diagnostic models in the medical field learn patterns for diseases from patient data. Voice assistants use machine learning to translate speech. Basically, any application that needs to make sense of data relies on machine learning.

    Deep learning – a sub-form of machine learning with deep neural networks – has achieved enormous leaps in performance in many areas (image and speech recognition in particular). The ability to generalize is always important in machine learning: the model should not only know the learned examples by heart, but also understand the rules behind them. If it does not do the latter, it is often overfitting, which is to be prevented by regular testing on unknown data and suitable techniques.
    Back to the overview