Reproducibility and Efficiency of Scientific Data Analysis: Scientific Workflows and Case-Based Reasoning

摘要

Scientists carry out complex scientific data analyses by managing and executing many related computational steps. Typically, scientists find a type of analysis relevant to their data, implement it step by step to try it out, and run many variants as they explore different datasets or method configurations. These processes are often done manually and are prone to error, slowing the pace of discoveries. Scientific workflows have emerged as a formalism to represent how the individual steps work and how they relate to the overall process. Workflows can be published, discovered, and reused to make data analysis processes more efficient through automation and assistance. In this talk, I will argue that integrating case-based reasoning techniques with workflows research would result in improved approaches to workflow sharing, retrieval, and adaptation. I will describe our initial work on semantic workflow matching using labeled graphs and knowledge intensive similarity measures. Furthermore, I will argue that if scientists followed a case-based approach more closely, scientific results would be more easily inspectable and reproducible. Through scientific workflows and case-based reasoning, scientific data analysis could be made more efficient and more rigorous.