Practically every computer program uses algorithms. An algorithm contains instructions that are to be followed step by step in order to achieve a particular goal. The instructions are found in the source code of software, i.e. the text written in a programming language and readable for humans. Algorithms are as diverse as the applications they enable: for example, the Shortest Path Algorithm calculates the fastest routes in navigation systems, the PageRank algorithm is used in generating the results of online search engines.
An area of computer science that deals with → machine learning and the implementation of intelligent behaviour using → algorithms. The goal is to replicate human cognitive abilities in computer systems. Artificial intelligence is used in many industries and areas that work with large volumes of data. Applications exist both in industry and in private households, including digital assistants (Siri, Alexa and navigation systems et al.), production support systems, mobility applications, planning tools in logistics, smart home applications and many more.
Model from the field of → AI that is based on the human brain and its network of neurons/nerve cells. The programme structure of the network includes data nodes that are networked in layers. The connections between the layers are weighted differently according to different training data and methods. Depending on the complexity of the neuron networking (number of layers), complex patterns can be detected and processed (→ Deep Learning).
Part of → Machine Learning, this describes the program structure of → artificial neural networks consisting of a complex inner program structure of several layers. The algorithms independently develop learning models by comparing, linking and drawing conclusions from different data on different layers. The machine learns to learn and to make decisions. Computing processes are more complex in deep learning and require more computing power than in the case of simple machine learning.
Generative Adversarial Networks Algorithm Model from the field of → Machine Learning that consists of two → artificial neural networks whose → algorithms perform conflicting learning tasks. The two neural networks play off against one another. Network 1 (generator) generates new data on the basis of the data provided; network 2 (discriminator) evaluates this data. By repeatedly comparing the results, the algorithm learns and always generates new data that network 2 can no longer distinguish from the true original data. Increasingly perfect “deceptions” or “fakes” are generated. Applications include, for example, the creation of photorealistic images (→ StyleGAN) and user-oriented communication with chatbots. The goal is to give users the impression of visual and communicative authenticity.
Face generator software based on a → GAN model for creating artificially assembled photorealistic images of human faces. The image of the artificial face is built up step by step on different layers, whereby the different features (shape, colour) are controlled individually. The results are high-resolution, authentic-looking images of people who do not exist.
The space in which data is temporarily stored after it is entered into an → AI system and before it is converted and outputted. In the → GAN model, the generator takes individual data information from the latent space and generates new data from it. In this hidden space, information such as image or audio data, which an AI has processed and learned, is mixed or even invented, creating new acoustic and visual worlds. The results can be abstract as well as concrete.
Method of artificial intelligence (→ AI) whereby systems automatically learn from training data and experiences and optimise themselves without having been previously programmed using defined rules or solutions. Machine Learning can be found in a wide variety of applications, from spam filters to personalised advertising and autonomous driving.
These determine, process and organise data into different categories based on patterns such as repetitions, similarities and rules. Typical applications include speech, text and facial recognition. These applications are found in facial recognition and fingerprint detection on mobile phones as well as in support systems for medical diagnoses.
A special → artificial neural network from the class of Convolutional Neural Networks that, among other things, is used for data processing of image and audio data in → AI technologies. The name VGG-16 stands for the developers Visual Geometry Group (VGG / University of Oxford) and the network structure, which consists of 16 convolutional layers.