In machine learning, we intuitively adopt an Observation-Oriented principle where observational variables act as the bedrock for relationships. It may suffice for conventional models, but with AI's capacities incorporating big data, it accentuates the misalignment between purely observational models and our actual comprehension. In contrast, humans construct cognitive entities indexed through relationships, which are not confined by observations, allowing us to formulate knowledge across temporal and hyper-dimensional spaces. This study introduces a novel Relation-Oriented perspective, drawing intuitive examples from computer vision and health informatics, to redefine our context of modeling with a causal focus. Furthermore, we present an implementation method - the relation-defined representation modeling, the feasibility of which is substantiated through comprehensive experiments.

翻译：在机器学习中，我们直觉性地采用面向观测的原则，即观测变量作为关系的基础。这对传统模型或许足够，但随着人工智能整合大数据的能力增强，这凸显了纯观测模型与我们实际理解之间的错位。相比之下，人类构建以关系索引的认知实体，这些实体不受观测约束，使我们能够在时间和超维空间中构建知识。本研究提出一种新颖的面向关系视角，通过计算机视觉和健康信息学中的直观示例，以因果焦点重新定义建模语境。此外，我们提出一种实现方法——关系定义表示建模，并通过全面实验验证其可行性。