Out-of-distribution detection is a research area and methodological approach within machine learning focused on identifying input data points that originate from a probability distribution significantly different from the one used to train a given model. This field investigates techniques to distinguish between data drawn from the known training distribution and novel or anomalous data, often assessing model uncertainty or confidence, with the primary significance being the enhancement of system robustness, safety, and reliability in real-world deployment scenarios by preventing potentially erroneous predictions on inputs for which the model is not adequately trained.