Training data versioning is an important concept related to artificial intelligence, big data and smart data, as well as automation. Data is the core of any AI application. For an AI to continuously improve, it must be constantly „trained“ with new, high-quality data. However, not every data collection remains the same – it changes and develops over time.
With training data versioning, every single state, every change or addition to the data is recorded. This works similarly to software development, where there are different versions of a program. This allows you to always trace which data an AI was trained with and how it has evolved over time.
A clear example: A company uses an AI that automatically recognises product images. Initially, only 5,000 images were used, later another 10,000 were added. Training data versioning allows us to see precisely when the additional images were included and how the recognition improved as a result. This ensures traceability, better control, and more trust in the AI's results.
Deutsch 
English (UK)