Visual Concept-Metaconcept Learning
This addresses the challenge of robust visual reasoning for AI systems, though it appears incremental in advancing concept learning methods.
The paper tackles the problem of jointly learning visual concepts and metaconcepts from images and question-answer pairs, proposing VCML to exploit their bidirectional connection for improved generalization and learning from limited data, with validation on synthetic and real-world datasets.
Humans reason with concepts and metaconcepts: we recognize red and green from visual input; we also understand that they describe the same property of objects (i.e., the color). In this paper, we propose the visual concept-metaconcept learner (VCML) for joint learning of concepts and metaconcepts from images and associated question-answer pairs. The key is to exploit the bidirectional connection between visual concepts and metaconcepts. Visual representations provide grounding cues for predicting relations between unseen pairs of concepts. Knowing that red and green describe the same property of objects, we generalize to the fact that cube and sphere also describe the same property of objects, since they both categorize the shape of objects. Meanwhile, knowledge about metaconcepts empowers visual concept learning from limited, noisy, and even biased data. From just a few examples of purple cubes we can understand a new color purple, which resembles the hue of the cubes instead of the shape of them. Evaluation on both synthetic and real-world datasets validates our claims.