Learning from examples to improve Automation in Model-Driven Engineering

模型驱动工程（MDE）由于能够降低软件开发和维护的复杂性，近年来受到了广泛的关注。然而，尽管在过去十年中报告了一些成功的故事，但MDE仍处于采用的早期阶段。采用MDE的一个主要障碍是很难自动化这种开发范式的许多活动。自动化是MDE的基石和基本原则。在这个范例中，领域特定的建模语言与转换引擎和生成器相结合，以产生各种软件工件。定义建模语言、编写转换以及维护所涉及的模型和其他工件之间的一致性是难以自动化的典型任务。这种困难主要来自于缺乏某些特定领域的知识。元模型和转换机制的定义基本上就是这种情况。自动化的困难也可以与一些手动任务引入的各种不一致性，如在模型maintenance.To帮助在MDE的自动化改进的情况下，我们建议使用的工件定义的输入和输出的任务自动化的例子。根据这项任务，示例可以采取不同的形式。例如，对于模型转换或重构，示例是成对的源模型和目标模型。类似地，在元模型定义中，示例可以是标记为有效/无效的模型。任务的自动化被视为一个优化过程，它可以导出最符合手头示例的自动化知识。根据问题的大小，即，搜索空间，这样的过程可以使用穷举或启发式搜索。这些任务包括模型转换、模型重构、转换规范和精确的元模型定义。

The model-driven engineering (MDE) received much attention in recent years due to its promise to reduce the complexity of the development and maintenance of software applications. However, and notwithstanding the success stories reported in the past decade, MDE is still at the early stages of adoption. One major obstacle to the adoption of MDE is the difficulty to automate many activities of this development paradigm.Automation is a keystone and a founding principle of MDE. In this paradigm, domain-specific modeling languages are combined with transformation engines and generators to produce various software artifacts. Defining modeling languages, writing transformations, and maintaining consistency between the involved models and other artifacts are typical tasks that are difficult to automate. This difficulty comes mainly from the lack of knowledge in some specific domains. This is essentially the case for the definition of meta-models and transformation mechanisms. Difficulty of automation can also be related to various inconsistencies introduced by some manual tasks as in the case of model maintenance.To help in the improvement of automation in MDE, we propose to use examples of artifacts that define the inputs and outputs of the task to automate. Depending on this task, examples could take different forms. For model transformation or refactoring, for instance, examples are pairs of source and target models. Similarly, in metamodel definition, examples could be models that are labeled as valid/invalid. The automation of a task is viewed as an optimization process that derives the automation knowledge that best conforms to the examples at hand. Depending on the size of the problem, i.e., search space, such a process could use an exhaustive or heuristic search.The above-mentioned idea will be explored for different MDE tasks. These tasks include model transformation, model refactoring, transformation specification, and precise metamodel definition.