知识卡片

普通读书笔记卡 · 7013.d · 生成模型:DeepSeek-V4.1-Flash

检测更难、参与者更广、时间线更模糊

理解等级:中级

在通用角色骨架之上,作者列出 ML 事件真正变化的三处。检测:ML 系统比非 ML 系统更不确定,因此很难写出能在人类用户察觉之前捕获全部故障的监控规则;未被良好埋点的系统常把基础设施问题只表现为质量缓慢劣化,模型质量变化往往是唯一的端到端信号。参与排障与恢复的角色与系统:ML 故障通常牵动更广的人员,包括业务/产品与管理层,因为 ML 系统建立在多个复杂系统之上、又影响多个系统,天然带来多元利益相关者。时间线与解决不清:许多 ML 事件影响的是本就随时间波动的质量指标,很难精确界定起止,也很难区分「修好了」与「持续改进」。这三条在后文被提炼为 public(用户先发现)、fuzzy(影响与时间双向模糊)、unbounded(跨系统跨组织)三条指导原则,也是为什么本章反复强调端到端质量指标、业务侧协作与提前约定宣布标准。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • Incident Response · 查看原书依据

      The things that do vary are listed here: Detection ML systems are less deterministic than non-ML systems. As a result, it is harder to write monitoring rules to catch all incidents before a human user detects them. Roles and systems involved in resolution ML incidents usually involve a broader range of staff during troubleshooting and resolution, including business/product and management/leadership.

    参考来源

    - 位置:《Reliable Machine Learning》第11章《Incident Response》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch11.xhtml) - 结论依据:原文明确阐述该主题,并给出定义、实践或反例。 - 原始内容:The things that do vary are listed here: Detection ML systems are less deterministic than non-ML systems. As a result, it is harder to write monitoring rules to catch all incidents before a human user detects them. Roles and systems involved in resolution ML incidents usually involve a broader range of staff during troubleshooting and resolution, including business/product and management/leadership.