Training metrics for AI are an important term in artificial intelligence, big data, smart data and digital transformation. They help to objectively evaluate and improve the learning progress and performance of AI models.
When developing an artificial intelligence, for example an image recognition programme, an algorithm is "trained" with many examples. Training metrics for AI then measure how well the programme masters its task. Common metrics include "accuracy", which measures how many images are recognised correctly, or "loss", which shows how much the result still deviates from the optimum.
Imagine you are training an AI to recognise cats and dogs in photos. Using the training metrics for AI, you can see, for example, that the model recognises 90 % of all cats correctly, but still makes mistakes with dogs. This allows developers to make targeted improvements and recognise when the model is "good enough" or needs further training.
Training metrics for AI are therefore a key tool for making the performance of artificial intelligence transparent and measurable.
Deutsch 
English (UK)