An Empirical Analysis of Backward Compatibility in Machine Learning Systems

被引:18
|
作者
Srivastava, Megha [1 ]
Nushi, Besmira [1 ]
Kamar, Ece [1 ]
Shah, Shital [1 ]
Horvitz, Eric [1 ]
机构
[1] Microsoft Res, Redmond, WA 98052 USA
关键词
D O I
10.1145/3394486.3403379
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
In many applications of machine learning (ML), updates are performed with the goal of enhancing model performance. However, current practices for updating models rely solely on isolated, aggregate performance analyses, overlooking important dependencies, expectations, and needs in real-world deployments. We consider how updates, intended to improve ML models, can introduce new errors that can significantly affect downstream systems and users. For example, updates in models used in cloud-based classification services, such as image recognition, can cause unexpected erroneous behavior in systems that make calls to the services. Prior work has shown the importance of "backward compatibility" for maintaining human trust. We study challenges with backward compatibility across different ML architectures and datasets, focusing on common settings including data shifts with structured noise and ML employed in inferential pipelines. Our results show that (i) compatibility issues arise even without data shift due to optimization stochasticity, (ii) training on large-scale noisy datasets often results in significant decreases in backward compatibility even when model accuracy increases, and (iii) distributions of incompatible points align with noise bias, motivating the need for compatibility aware de-noising and robustness methods.
引用
收藏
页码:3272 / 3280
页数:9
相关论文
共 50 条
  • [41] Machine Learning for Systems
    Litz, Heiner
    Hashemi, Milad
    IEEE MICRO, 2020, 40 (05) : 6 - 7
  • [42] An Empirical Evaluation of Intelligent Machine Learning Algorithms under Big Data Processing Systems
    Suleiman, Dima
    Al-Zewairi, Malek
    Naymat, Ghazi
    8TH INTERNATIONAL CONFERENCE ON EMERGING UBIQUITOUS SYSTEMS AND PERVASIVE NETWORKS (EUSPN 2017) / 7TH INTERNATIONAL CONFERENCE ON CURRENT AND FUTURE TRENDS OF INFORMATION AND COMMUNICATION TECHNOLOGIES IN HEALTHCARE (ICTH-2017) / AFFILIATED WORKSHOPS, 2017, 113 : 539 - 544
  • [43] Empirical Architecture Comparison of Two-input Machine Learning Systems for Vision Tasks
    Wakigami, Kazuya
    Machida, Fumio
    Phung-Duc, Tuan
    Formal Aspects of Computing, 2024, 36 (04)
  • [44] On the compatibility of established methods with emerging artificial intelligence and machine learning methods for disaster risk analysis
    Thekdi, Shital
    Tatar, Unal
    Santos, Joost
    Chatterjee, Samrat
    RISK ANALYSIS, 2024,
  • [45] An empirical analysis on webservice antipattern prediction in different variants of machine learning perspective
    Lov Kumar
    Sahiti Tummalapalli
    Lalita Bhanu Murthy
    Sanjay Misra
    Aneesh Krishna
    Scientific Reports, 15 (1)
  • [46] An empirical analysis of homicides in Mexico through Machine Learning and statistical design of experiments
    Silva Urrutia, Jose Eliud
    Villalobos, Miguel A.
    POBLACION Y SALUD EN MESOAMERICA, 2022, 20 (01):
  • [47] An empirical assessment of machine learning approaches for triaging reports of static analysis tools
    Sai Yerramreddy
    Austin Mordahl
    Ugur Koc
    Shiyi Wei
    Jeffrey S. Foster
    Marine Carpuat
    Adam A. Porter
    Empirical Software Engineering, 2023, 28
  • [48] Incorporating Empirical Orthogonal Function Analysis into Machine Learning Models for Streamflow Prediction
    Wu, Yajie
    Chen, Yuan
    Tian, Yong
    SUSTAINABILITY, 2022, 14 (11)
  • [49] An Empirical Comparative Analysis Using Machine Learning Techniques for Liver Disease Prediction
    Alghobiri, Mohammed
    Khan, Hikmat Ullah
    Mahmood, Ahsan
    INTERNATIONAL JOURNAL OF HEALTHCARE INFORMATION SYSTEMS AND INFORMATICS, 2021, 16 (04)
  • [50] To Seed or Not to Seed? An Empirical Analysis of Usage of Seeds for Testing in Machine Learning Projects
    Dutta, Saikat
    Arunachalam, Anshul
    Misailovic, Sasa
    2022 IEEE 15TH INTERNATIONAL CONFERENCE ON SOFTWARE TESTING, VERIFICATION AND VALIDATION (ICST 2022), 2022, : 151 - 161