Causal Reasoning Meets Visual Representation Learning: A Prospective Study

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