知识卡片

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

三条指导原则与四类角色的准备清单

理解等级:中级

抽象三个故事,作者给出三条反复出现的原则:面向公众的 ML 故障常最先被终端用户或管线最末端发现,因为质量监控难,只影响一小撮用户全部时间的问题在聚合指标上完全看不出来;ML 故障在时间与影响上都模糊,起止难定,「坏了」和「还能更好」之间没有清晰分界;排障与恢复不受边界约束,牵动组织的技术、产品、业务面更广。按角色的准备随之展开:模型/数据人员版本化全部模型与数据(含转换代码与可直接服务的历史模型)、指定可接受回退、选定与实现无关的质量指标,并在事件中保持可联络、准备做定制分析,同时敢于拒绝违反隐私伦理的请求;软件工程师保证数据血缘清晰、训练与服务特征一致以防 training-serving skew、模型与二进制发布解耦、备好回滚与版本查询工具;ML SRE 预留余量、定期演练与架构评审、设 SLO/SLI 并主动理解业务;业务负责人理解技术能力与局限、保障组织事件容量与复盘完成标准。

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

      Guiding Principles Three overarching themes that appear across ML incidents are so common that we list them here as guiding principles: Public ML outages are often detected first by end users, or at least at the very end of the pipeline, all the way out in serving or integrated into an application. This is partly true because ML model performance (quality) monitoring is difficult.

    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:Guiding Principles Three overarching themes that appear across ML incidents are so common that we list them here as guiding principles: Public ML outages are often detected first by end users, or at least at the very end of the pipeline, all the way out in serving or integrated into an application. This is partly true because ML model performance (quality) monitoring is difficult.