Abstract Reasoning Tasks (ART)

The Abstract Reasoning Tasks (ART) dataset was proposed by Webb et al. [51], consisting of four visual reasoning tasks, each defined by a different abstract rule. In the ‘same/different’ task, two objects are presented, and the task is to say whether they are the same or different. In the ‘relational match-to-sample’ task, a source pair of objects is presented that either instantiates a ‘same’ or ‘different’ relation, and the task is to select the pair of target objects that instantiates the same relation. In the ‘distribution-of-3’ task, a set of three objects is presented in the first row, and an incomplete set is presented in the second row. The task is to select the missing object from a set of four choices. In the ‘identity rules’ task, an abstract pattern is instantiated in the first row (ABA, ABB, or AAA), and the task is to select the choice that would result in the same relation being instantiated in the second row.