Models in the Ontology（本体论中的模型）¶

Organizations are looking to leverage artificial intelligence (AI) and machine learning (ML) to accelerate and improve decision-making. But the reality of operationalizing AI/ML is complex, and the typical return on investment rarely lives up to expectations.

Foundry provides the key capabilities necessary to bridge this gap: a trustworthy data foundation, tools for evaluating and comparing models against organizational objectives, and functionality for deploying models into user-facing operational workflows. This page focuses on the last step: deploying an evaluated model into production.

End-to-end workflow¶

At a high level, these are the end-to-end steps required to operationalize AI/ML in Foundry for live inference with the Ontology:

Create a model in Foundry.
Configure a direct model deployment.
Publish a simple wrapper function for your model and optionally call it from another function to orchestrate complex logic around your model.
Use that function for live inference in Workshop, Vertex and other end-user facing applications.

Ontology Objects can also be backed with datasets that leverage a model for batch inference - learn how to use a model in Code Repositories.

Benefits¶

Just like mapping datasets to Ontology concepts provides benefits for workflow development and decision-making, mapping models to the Ontology provides a number of benefits:

Interpretability. Because all modeling results are defined in terms of real-world concepts (properties of an object type), end users do not need to understand machine learning in order to use modeling results. Instead, users simply interact with simple concepts such as a forecast, estimate, or classification.
Economies of scale. Instead of each modeling project being a bespoke effort created in service of a specific use case, modeling efforts can build on each other over time. For example, a forecast produced for one use case can immediately be used for subsequent use cases as well, reducing duplicated effort and providing end-user value more quickly over time.
Connectivity at scale. By incorporating ML models, the Ontology becomes a single source of truth for the organization, not just in terms of data, but also in terms of logic. Models encode the organization's expectations for how things may change in the future. In this way, the Ontology becomes a "digital twin" for the entire enterprise, which unlocks the ability to simulate changes across the organization in ways that would never be possible with a wide array of disparate modeling efforts.

中文翻译¶

本体论中的模型¶

各类组织正寻求利用人工智能(AI)和机器学习(ML)来加速并改进决策过程。然而，将AI/ML投入实际运营的复杂性远超想象，典型的投资回报率往往难以达到预期。

Foundry提供了弥合这一差距所需的关键能力：可信的数据基础、评估和比较模型与组织目标的工具，以及将模型部署到面向用户的操作工作流中的功能。本页重点介绍最后一步：将经过评估的模型部署到生产环境中。

端到端工作流¶

从宏观角度来看，以下是在Foundry中利用本体论(Ontology)实现AI/ML实时推理所需的端到端步骤：

在Foundry中创建模型
配置直接模型部署
为模型发布简单的包装函数，并可选地从其他函数调用它，以编排围绕模型的复杂逻辑
在Workshop、Vertex及其他面向最终用户的应用程序中使用该函数进行实时推理

本体论对象也可以由利用模型进行批量推理的数据集支持——了解如何在代码仓库中使用模型

优势¶

正如将数据集映射到本体论概念能为工作流开发和决策带来优势一样，将模型映射到本体论也能带来诸多好处：

可解释性(Interpretability)。由于所有建模结果都以现实世界概念（对象类型的属性）来定义，最终用户无需理解机器学习即可使用建模结果。用户只需与诸如预测值、估计值或分类结果等简单概念进行交互。
规模经济(Economies of scale)。建模工作不再是针对特定用例的定制化项目，而是可以随着时间的推移相互叠加。例如，为某个用例生成的预测结果可以立即用于后续用例，从而减少重复工作，并更快地为最终用户创造价值。
规模化连接(Connectivity at scale)。通过整合ML模型，本体论成为组织的单一事实来源——不仅体现在数据层面，更体现在逻辑层面。模型编码了组织对未来变化的预期。通过这种方式，本体论成为整个企业的"数字孪生"，能够解锁跨组织模拟变化的能力，这是分散的建模工作永远无法实现的。