2021/03/25 by Grace A. Lewis, Lewis, Grace A., Stephany Bellomo +3 · 1 citation
Business, Management and Accounting · Computer Science · Decision Sciences · #Artificial Intelligence (cs.AI) #Big Data and Business Intelligence #Data Quality and Management #FOS: Computer and information sciences #Machine Learning (cs.LG) #Software Engineering (cs.SE) #Software Engineering Research
paper · pdf · doi:10.48550/arxiv.2103.14101
openalex publication_date 2021/03/25 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Increasing availability of machine learning (ML) frameworks and tools, as\nwell as their promise to improve solutions to data-driven decision problems,\nhas resulted in popularity of using ML techniques in software systems. However,\nend-to-end development of ML-enabled systems, as well as their seamless\ndeployment and operations, remain a challenge. One reason is that development\nand deployment of ML-enabled systems involves three distinct workflows,\nperspectives, and roles, which include data science, software engineering, and\noperations. These three distinct perspectives, when misaligned due to incorrect\nassumptions, cause ML mismatches which can result in failed systems. We\nconducted an interview and survey study where we collected and validated common\ntypes of mismatches that occur in end-to-end development of ML-enabled systems.\nOur analysis shows that how each role prioritizes the importance of relevant\nmismatches varies, potentially contributing to these mismatched assumptions. In\naddition, the mismatch categories we identified can be specified as machine\nreadable descriptors contributing to improved ML-enabled system development. In\nthis paper, we report our findings and their implications for improving\nend-to-end ML-enabled system development.\n