知识卡片
流程一:关键干系人的定期评审
引入 ML 对组织的影响往往是弥漫性的,为缓解集中化的缺点,可以用流程把决策或知识分发(即去中心化)出去。第一类是关键干系人评审:由 ML 团队定期汇报,目的有两层——确保当前建模结果获得批准,以及让业务领导者理解系统正在如何适应业务。作者特别提醒「评审该纳入哪些指标」本身就是一门需要单独研究的学问:指标必须完整,既包含模型实现带来的改进,也包含它的成本;关键业务干系人还可以复核 ML 团队各项工作的优先级、ROI 与用例。这类评审的价值在于把「模型好不好」从建模团队的内部判断,变成业务方共同签字确认的组织判断,正好补上集中式结构里「业务单元不理解可靠性流程」的缺口。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
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
- Practical ML Org Implementation Examples · 查看原书依据
These processes include the following: Reviews by key stakeholders These reviews, presented by the ML team on a regular basis, should ensure approval of the current modeling results as well as an understanding by business leaders of the way the system is adapting to the business. A separate science of exactly which metrics to include in these reviews is necessary, but the metrics should be complete and need to include the impr