Visual Superordinate Abstraction for Robust Concept Learning  

作　　者：Qi Zheng Chao-Yue Wang Dadong Wang Da-Cheng Tao 

机构地区：[1]University of Sydney,Sydney 2008,Australia [2]JD Explore Academy,Beijing 100176,China [3]DATA61,Commonwealth Scientific and Industrial Research Organisation,Sydney 2122,Australia

出　　处：《Machine Intelligence Research》2023年第1期79-91,共13页机器智能研究（英文版）

基　　金：supported in part by the Australian Research Council(ARC)(Nos.FL-170100117,DP-180103424,IC-190100031 and LE-200100049).

摘　　要：Concept learning constructs visual representations that are connected to linguistic semantics, which is fundamental to vision-language tasks. Although promising progress has been made, existing concept learners are still vulnerable to attribute perturbations and out-of-distribution compositions during inference. We ascribe the bottleneck to a failure to explore the intrinsic semantic hierarchy of visual concepts, e.g., {red, blue,···} ∈“color” subspace yet cube ∈“shape”. In this paper, we propose a visual superordinate abstraction framework for explicitly modeling semantic-aware visual subspaces(i.e., visual superordinates). With only natural visual question answering data, our model first acquires the semantic hierarchy from a linguistic view and then explores mutually exclusive visual superordinates under the guidance of linguistic hierarchy. In addition, a quasi-center visual concept clustering and superordinate shortcut learning schemes are proposed to enhance the discrimination and independence of concepts within each visual superordinate. Experiments demonstrate the superiority of the proposed framework under diverse settings, which increases the overall answering accuracy relatively by 7.5% for reasoning with perturbations and 15.6% for compositional generalization tests.

关 键 词：Concept learning visual question answering weakly-supervised learning multi-modal learning curriculum learning 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]