The term "verification of neural networks" is particularly at home in the fields of artificial intelligence, automation and cybersecurity. It describes methods and processes that are used to check whether artificial intelligence - i.e. neural networks - actually do what they are supposed to do.
Neural networks are used in autonomous driving, for example, to recognise traffic signs or pedestrians. Verification checks whether the system works safely and also reacts correctly in unusual situations. This prevents the AI from making mistakes that could lead to dangerous situations.
A simple example: a neural network should always recognise cats in photos. Verification checks whether the AI really recognises all possible cats, even if they look different or the background changes. It also checks that the system does not accidentally mistake dogs for cats.
Verification of neural networks is particularly important in safety-critical applications in order to create trust. Companies and users want to be sure that the AI works reliably and that unintentional errors are ruled out. This is why the verification of neural networks is becoming increasingly important the more intelligent systems are used in our everyday lives.
Deutsch 
English (UK)