2. Detection and correction by machine learning

Recent work has shown that machine learning models can be used to accurately identify problems in data and correct certain types of error with complex (semi-)automatic correction mechanisms that were previously performed manually or based on hard-to-maintain heuristics. New learning strategies can also determine which corrections are necessary depending on the analysis objective, as it is not necessarily necessary (or even feasible) to correct all problems.

Learning-based approaches can be used to detect or correct erroneous data. They rely on examples of erroneous and correct records to train the model. But designing models that are sufficiently expressive and therefore complex requires the use of a large number of examples. Depending on the task (detection of outliers, duplicates, inconsistencies, etc.), the creation of these sets of examples can prove very difficult,...