Visual representation learning is ubiquitous in various real-world applications,including visual comprehension,video un-derstanding,multi-modal analysis,human-computer interaction,and urban computing.Due to the emergence of huge amounts of multi-modal heterogeneous spatial/temporal/spatial-temporal data in the big data era,the lack of interpretability,robustness,and out-of-dis-tribution generalization are becoming the challenges of the existing visual models.The majority of the existing methods tend to fit the original data/variable distributions and ignore the essential causal relations behind the multi-modal knowledge,which lacks unified guidance and analysis about why modern visual representation learning methods easily collapse into data bias and have limited general-ization and cognitive abilities.Inspired by the strong inference ability of human-level agents,recent years have therefore witnessed great effort in developing causal reasoning paradigms to realize robust representation and model learning with good cognitive ability.In this paper,we conduct a comprehensive review of existing causal reasoning methods for visual representation learning,covering fundament-al theories,models,and datasets.The limitations of current methods and datasets are also discussed.Moreover,we propose some pro-spective challenges,opportunities,and future research directions for benchmarking causal reasoning algorithms in visual representation learning.This paper aims to provide a comprehensive overview of this emerging field,attract attention,encourage discussions,bring to the forefront the urgency of developing novel causal reasoning methods,publicly available benchmarks,and consensus-building stand-ards for reliable visual representation learning and related real-world applications more efficiently.

Yang Liu;Yu-Shen Wei;Hong Yan;Guan-Bin Li;Liang Lin

School of Computer Science and Engineering,Sun Yat-sen University,Guangzhou 510006,China

机器智能研究（英文）

[1]Yang Liu;Yu-Shen Wei;Hong Yan;Guan-Bin Li;Liang Lin-.Causal Reasoning Meets Visual Representation Learning:A Prospective Study)[J].机器智能研究（英文）,2022(06):485-511

fundament

Causal,Reasoning,Meets,Visual,Representation,Learning,Prospective,Study,representation,learning,ubiquitous,various,world,applications,including,visual,comprehension,video,derstanding,multi,modal,analysis,human,computer,interaction,urban,computing,Due,emergence,huge,amounts,heterogeneous,spatial,temporal,big,interpretability,robustness,generalization,are,becoming,challenges,existing,models,majority,methods,tend,fit,original,variable,distributions,ignore,essential,causal,relations,behind,knowledge,which,lacks,unified,guidance,about,why,modern,easily,collapse,into,bias,have,limited,cognitive,abilities,Inspired,by,strong,inference,level,agents,recent,years,therefore,witnessed,great,effort,developing,reasoning,paradigms,realize,good,this,paper,we,conduct,comprehensive,review,covering,theories,datasets,limitations,current,also,discussed,Moreover,propose,some,opportunities,future,research,directions,benchmarking,algorithms,This,aims,provide,overview,emerging,field,attract,attention,encourage,discussions,bring,forefront,urgency,novel,publicly,available,benchmarks,consensus,building,ards,reliable,related,more,efficiently

0.602351