What are the requirements for machine learning

What is machine learning?

Artificial intelligence has supported us in business, research and development for many years. Machine learning is one of the most important and successful disciplines in artificial intelligence. But what is behind it?

Definition: what is machine learning?

Machine learning (ML) is aPart of Artificial Intelligence and thus a form of applied mathematics and computer science. With machine learning, IT systems can use data independentlyGenerate knowledge,Build algorithms, learn automatically and recognize new connections. The aim is to apply identified patterns to a new data set and such aOptimizing the results or betterMake predictions to be able to.

How does machine learning work?

People serve as the basis for the process outlined abovePredefined training data and manually edited featureswhich one must pay attention to. So “only” the assignment of these characteristics is learned. This is done viaAlgorithms built a static model. This recognizes recurring patterns and saves them in parameters. The computer learns in the same way as we humans doindependently from experience and generates new knowledge in the process.

The three most important algorithmic approaches in machine learning

  • Supervised learning
  • Unsupervised learning (unsupervised learning)
  • Reinforcement Learning

Supervised learning is probably the simplest method of machine learning. With this form, the algorithm already knows the data and the desired result. Often this data is also calledTraining and test data set designated. Using these examples, the algorithm gets to know the logic behind it. He is then able to classify data that have a certain similarity to the training set according to the learned logic. The learning takes place here "monitored".

At theUnsupervised learning however areneither target values ​​nor structures given by data. The aim of unsupervised learning is to independently recognize structures and thus therelevant information to filter out within the data. This type of machine learning is the most demanding.

The third and currently increasingly important type of machine learning is thatReinforcement learning. The algorithm learns through thisPunishment or reward. The system is set up in such a way that it is rewarded for successful behavior, e.g. B. when it achieves a goal. The system will be punished for behavior that leads to an undesirable situation. In this case the algorithm has to start over. So the system learns based on thatFeedback from his environment. This form is similar to human learning. You can see a nice example of what (deep) reinforcement learning can do in the form of this parking training:

By loading the video, you accept YouTube's privacy policy.
Learn more

Load video

What do you need machine learning for?

In times of huge floods of information and data, such IT systems have become irreplaceable. They help companies get out of Big DataWhat is Big Data? The term “Big Data” comes from the English-speaking world and stands for large data or mass data. It denotes amounts of data that are • too large, • ... read more Smart data to generate. With the help of artificial intelligence and machine learning, unmanageable amounts of data for the human brain can be analyzed and evaluated without our intervention, or simply structured and ordered. Alsoin everyday life We already encounter machine learning - albeit mostly unconsciously. Systems equipped with this should enable us to work more effectively and act more comfortably. In technology and industry, machine learning is astrong innovation driver. Since the algorithms are able tocomplex tasks totake, they are common atMonitoring of processes used. So you can z. B. Control individual process steps and identify potential problems or damage at an early stage (Predictive maintenance). Machine learning is also used in medicine and supports doctors and scientists, e.g. B. successful in theCancer research.

The following list of other areas of application and software applications should enable you to better assess the importance of machine learning algorithms:

  • Speech recognition and digital assistants
  • Movie recommendations on Netflix
  • Book recommendations on Amazon
  • Face recognition from Facebook
  • Google ranking
  • Data analysis and data management
  • find the most effective therapies for patients
  • CRM (Customer Relationship Management) systems
  • Image data analysis (CT / MRT) for disease diagnosis
  • Cyber ​​security and data security - e.g. B. through spam filters
  • Credit card fraud detection
  • Regression analysis
  • personalized content
  • Weather forecasts

We relevance makers also use machine learning to generate optimal visibility on Google with the content created for our customers. How AI helps us e.g. B. in analyzing and evaluating millions of freely available content from the web and thereby identifying those topics that are highly relevant for the customer and his target group (s).

Definition of terms and connections: Artificial Intelligence, Artificial Intelligence, Semantics, Machine Learning and Deep Learning, Neural Networks, Predictive Analytics

If we deal with machine learning, we inevitably stumble across various terms. It is often unclear what is actually meant. We try to show differences and connections.

Artificial intelligence(AI) is only the German translation ofArtificial Intelligence: Definition Artificial Intelligence (AI) or Artificial Intelligence (KI) refers to the simulation of human intelligence in machines. Computers are programmed to have properties that are compatible with cognitive intelligence such as ... Read More (AI). As already mentioned, machine learning is a sub-area of ​​artificial intelligence. Artificial intelligence is used nowadays as a collective term to describe allIntelligence solutions or applications as a whole. These include B. also robotics, machine translation and much more.

Semantics definition: what is semantics? The term semantics comes from the Greek and means something like denote or belonging to the sign. Semantics is a branch of linguistics and is also ... Read more is a branch oflinguistics and describes the relationship of linguistic signs and their meaning. Signs can be words or parts of words, but also symbols and entire sentences. Combined withDisambiguation we learn z. For example, whether a post about lobsters deals with the crustacean or the make of the car. So semantics are used to achieve theMeaning of an object to classify and identify. Similar to machine learning, properties of an object are assigned according to corresponding patterns. In contrast to machine learning, however, hereno predictions to be hit. The semantics work statically and do not follow a learning process.

Deep learning again counts asPart of machine learnings and is often referred to as thatFurther development designated. A prerequisite for successful deep learning are large amounts of data and - on the technical side - so-calledNeural Networks. A neural network is an artificial, abstract model of our brain, made up of artificial neurons. In contrast to machine learning, in deep learning humans neither intervene in data analysis nor in decision-making processes - they only provide the necessary information (data). Deep learning independently develops new data models and can - completely without manual adjustment - link what has been learned with new content, draw conclusions from it and e.g. B. Enabling machines to make decisions independently.

ForPredictive Analytics machine learning is used. Thanks to increasing digitalization, this type of use is becoming more and more important. It is intended to predict future events and is used in many business areas, for example in meteorology, in online advertising (advertising effect) or in the form of forecasts.

How can you develop applications with machine learning support?

Artificial Intelligence (AI) with all its sub-areas and possible uses will remain a trend for many years to come. If you want to familiarize yourself with this in depth, you have a large selection of tools, because many providers now provide their own cloud services for machine learning. From IBM with Watson to Amazon Machine Learning and Microsoft with Azure ML Studio to Google Tensorflow, Apache Spark and a wide range of open source software: Even developers with little experience in machine learning are enabled to implement their own intelligent applications.

Friederike Scholz

As Head of Product, Frieda takes care of the further development and networking of the tools we use. She mainly brings her expertise to bear in articles on online advertising. In addition, Frieda's great anecdotes are an enrichment for every lunch break.

learn more ...

That might be interesting too ...