Exploring the spatial reasoning ability of neural models in human IQ tests

Although neural models have performed impressively well on various tasks such as image recognition and question answering, their reasoning ability has been measured in only few studies. In this work, we focus spatial explore the understanding of models. First, describe following two IQ tests: rotation shape composition. Using well-defined rules, constructed datasets that consist complexity levels. We designed a variety experiments terms generalization, evaluated six different baseline newly generated datasets. provide an analysis results factors affect generalization abilities Also, analyze how solve tests with visual aids. Our findings would valuable insights into machine difference between human.