AI Dictionary › AI Fundamentals
Fuori distribuzione (Out-of-Distribution)
Out-of-distribution refers to an input that differs significantly from the data an AI model was trained on, in content, format or context. On such inputs, a model's performance is generally less reliable, since the system faces patterns it never encountered during training. Recognizing these cases matters just as much as measuring accuracy on familiar data.
To evaluate out-of-distribution behavior, dedicated test sets are built with inputs deliberately different from typical ones, such as new domains, less-represented languages or unusual formats. Changes in the model's accuracy and reliability relative to in-distribution data are then measured, often also observing whether the model signals its own uncertainty instead of answering with excessive confidence. Good out-of-distribution behavior includes the ability to recognize its own limits.
It is a central concern for applications where real-world inputs can vary widely from training data, such as analyzing domain-specific documents or use in new markets and languages. It also helps decide when a system should decline to answer or request human intervention.
The concept originates in statistics and classical machine learning, where the distinction between in-distribution and out-of-distribution data has long been used to assess the generalization ability of predictive models.
From our network
HSE Genius: AI for Safety Data Sheets
Extract SDS data, H phrases and ECHA compliance checks in seconds, powered by AI.
Visit hsegenius.com →From the Agora Intelligence blog
📱 Download the Android app (beta) iOS coming soon
Say what you mean. Get what you need.
Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.