Process Conformance under Incomplete Information

流程感知信息系统协调一组基本操作的执行以实现业务目标，其中操作可以像函数调用一样细粒度，也可以像复杂业务事务一样粗粒度。这类系统的行为通常由过程模型来描述。然而，一旦在运行时记录了数据(通常是以日志或事件流的形式)，一致性问题就出现了：系统的建模行为及其记录的行为如何相互关联？回答这个问题是检测、解释和补偿面向过程的系统模型与其实际执行之间的任何偏差的基础。在流程自动化、数据传感和流程相关资源的大规模检测等趋势的推动下，当今世界的事件数据量和生成频率都在增加。因此，基于进程执行的完整历史的事后一致性检查不再是可行的选择。PROCI项目旨在为符合性检查提供基础，这种检查打破了全面访问事件数据的无处不在的假设，并能够在不完全信息下进行推理。具体地说，将开发用于抽样和在线一致性检查的模型和算法。考虑到常见一致性检查技术的指数时间复杂性，前者通过只考虑大量事件数据的一小部分来显著提高运行时性能。后者的目标是在事件流而不是静态日志上实现一致性检查时的空间效率问题。在任何一种情况下，一致性检查都将是有损的，基于流程的事件数据的不完整视图，并且需要分别使用日志或事件流的批的每个新样本来更新偏差的描述。因此，研究的中心挑战将是(1)设计用于推理部分一致性结果的形式基础，(2)设计用于抽样和在线一致性检查的算法，对数据不完整性引起的偏差提供统计保证，以及(3)针对底层数据分布的变化实现健壮性。将进行实验验证，以经验地展示与假定完全访问事件数据的最先进算法相比，在运行时间和空间效率方面取得的改进。

Process-aware information systems coordinate the execution of a set of elementary actions to reach a business goal, where actions may be as fine-grained as function calls or as coarse-grained as complex business transactions. The behaviour of such systems is commonly described by process models. However, once data is recorded during runtime, typically in the form of logs or streams of events, the question of conformance emerges: how do the modelled behaviour of a system and its recorded behaviour relate to each other? Answering this question is the basis for the detection, interpretation, and compensation of any deviation between a model of a process-oriented system and its actual execution. Driven by trends such as process automation, data sensing, and large-scale instrumentation of process-related resources, the volume of event data and the frequency at which it is generated is increasing in today's world. Post-mortem conformance checking based on a complete history of a process' execution is thus no longer a viable option. The ProCI project sets out to provide the foundations for conformance checking that breaks with the omnipresent assumption of comprehensive access to event data and enables reasoning under incomplete information. Specifically, models and algorithms will be developed for sampled and online conformance checking. Given the exponential time complexity of common conformance checking techniques, the former strives for drastic improvements in runtime by considering only a fraction of large volumes of event data. The latter targets the question of space efficiency when conformance checking is realised over streams of events rather than static logs. In either case, conformance checking will be lossy, grounded on an incomplete view on the event data of a process, and needs to update a description of deviations with each new sample of a log or batch of an event stream, respectively. The central research challenges therefore will be (1) to devise the formal foundations for reasoning about partial conformance results, (2) to devise algorithms for sampled and online conformance checking, giving statistical guarantees on the bias induced by data incompleteness, and (3) to achieve robustness against changes in underlying data distributions. Experimental validations will be performed to empirically demonstrate the achieved improvements in runtime and space efficiency compared to state-of-the-art algorithms that postulate complete access to event data.