目录
《2021 enterprise trends in machine learning 2021年机器学习的企业趋势》翻译与解读
Trend 1: AI/ML priority and budgets are increasing趋势 1：AI/ML 优先级和预算正在增加
Trend 2: Customer experience, automation take priority趋势二：客户体验、自动化优先
Trend 3: AI/ML “have” and “have-not” gap趋势三：AI/ML“有”和“无”的差距
Trend 4: Governance is top challenge by far趋势 4：治理是迄今为止最大的挑战
Trend 6: Need for organizational alignment趋势 6：需要组织协调
Trend 7: Maturity limited by alignment issues趋势 7：成熟度受限于对齐问题
Technical debt is piling up技术债务堆积如山
Trend 8: Deployment time is increasing趋势 8：部署时间正在增加
Trend 9: Data scientists spend too much time on deployment趋势 9：数据科学家在部署上花费过多时间
Trend 10: Improved outcomes with MLOps solutions趋势 10：使用 MLOps 解决方案改善结果
相关文章
AI：Algorithmia《2020 state of enterprise machine learning—2020年企业机器学习状况》翻译与解读
AI：Algorithmia《2021 enterprise trends in machine learning 2021年机器学习的企业趋势》翻译与解读
《2021 enterprise trends in machine learning 2021年机器学习的企业趋势》翻译与解读
作者：Diego Oppenheimer，首席执行官
平台：algorithmia.com | @algorithmia
简介：迭戈·奥本海默(Diego Oppenheimer)是Algorithmia的联合创始人兼首席执行官。此前，他设计、管理和发布了一些微软最常用的数据分析产品，包括Excel、Power Pivot、SQL Server和Power BI。他持有卡内基梅隆大学(Carnegie Mellon University)的信息系统学士学位和商业智能和数据分析硕士学位。
Methodology
Random sampling of 403 respondents Business leaders involved in machine learning initiatives at their organizations Organizations with $100M or more in revenue Independent third-party |
随机抽样 403 名受访者 在其组织中参与机器学习计划的商业领袖 收入超过 1 亿美元的组织 独立第三方 |
Report themes
Priority shifts Organizations are responding to economic uncertainty by dramatically increasing AI/ML investment Challenges remain Enterprises continue to face basic challenges across the ML lifecycle Technical debt is piling up Despite the increased investment, organizations are spending even more time and resources on model deployment MLOps preferences Organizations report improved outcomes with third-party MLOps solutions |
优先级转变 企业组织正在通过大幅增加 AI/ML 投资来应对经济不确定性 挑战依然存在 企业在整个机器学习生命周期中继续面临基本挑战 技术债务堆积如山 尽管增加了投资，但组织在模型部署上花费了更多的时间和资源 MLOps偏好 组织报告使用第三方 MLOps 解决方案改善了结果 |
10 trends
Priority shifts 1. AI/ML priority and budgets are increasing 2. Customer experience and automation are top-priority use cases for AI/ML 3. There’s a gap between AI/ML “haves” and “have-nots” |
优先级转变 1. AI/ML 优先级和预算正在增加 2. 客户体验和自动化是 AI/ML 的首要用例 3. AI/ML“有”和“无”之间存在差距 |
Challenges remain 4. Governance is the top challenge by far 5. Basic integration and compatibility issues remain 6. AI/ML success requires organizational alignment 7. Companies lack organizational alignment maturity |
挑战依然存在 4. 治理是迄今为止最大的挑战 5. 基本的集成和兼容性问题仍然存在 6. AI/ML 的成功需要组织协调 7. 公司缺乏组织一致性成熟度 |
Technical debt is piling up 8. Deployment time is increasing 9. Data scientists spend too much time on deployment |
技术债务堆积如山 8. 部署时间越来越长 9. 数据科学家在部署上花费太多时间 |
MLOps preferences 10. Organizations report improved outcomes with MLOps solutions |
MLOps偏好 10. 组织报告使用 MLOps 解决方案改善了结果 |
2021: The year of ML
This was a year of increased urgency for ML—but initial efforts to scale only created more technical debt. In 2021, organizations that invest in MLOps will reap the greatest rewards. |
今年是机器学习紧迫性增加的一年——但最初的扩展努力只会产生更多的技术债务。 2021 年，投资于 MLOps 的组织将获得最大的回报。 |
Priority shifts优先级转变
Trend 1: AI/ML priority and budgets are increasing趋势 1：AI/ML 优先级和预算正在增加
83% of organizations have increased A/ML budgets year-on-year
83% 的组织增加了 A/ML 预算
The average number of data scientists employed has increased76% year-on-year
数据科学家的平均就业人数同比增长了76%
Trend 2: Customer experience, automation take priority趋势二：客户体验、自动化优先
The percentage of respondents who selected more than five use cases for AI/ML in our survey increased 74% year-on-year
Customer experience and process automation represent the top AI/ML use cases
在我们的调查中，选择超过 5 个 AI/ML 用例的受访者比例同比增长 74%
客户体验和流程自动化代表了顶级 AI/ML 用例
For nearly all use cases,50% or more of organizations are increasing their usage of AI/ML
对于几乎所有用例，50% 或更多的组织正在增加对 AI/ML 的使用
Trend 3: AI/ML “have” and “have-not” gap趋势三：AI/ML“有”和“无”的差距
More than half of all respondents have more than 25 models in production.
40% of all respondents have more than 50 models in production.
超过一半的受访者拥有超过 25 款型号在生产中。
40% 的受访者拥有 50 多款型号在生产中。
The world's largest enterprises are dominating the high end of model scale
世界上最大的企业正在主导模型规模的高端
Challenges remain挑战依然存在
Trend 4: Governance is top challenge by far趋势 4：治理是迄今为止最大的挑战
56% of organizations struggle with governance, security, andauditability issues |
56% 的组织在治理、安全和可审计性问题上苦苦挣扎 |
When asked about regulations they need to comply with for AI/ML, 67% of respondents selected multiple regulations. Only 8% selected no regulations at all. 67% of all organizations must comply with multiple regulations. |
当被问及他们需要遵守的 AI/ML 法规时，67% 的受访者选择了多项法规。 只有 8% 的人根本没有选择任何法规。 67% 的组织必须遵守多项法规。 |
Trend 6: Need for organizational alignment趋势 6：需要组织协调
Successful AI/ML initiatives involve decision-makers from across the organization
成功的 AI/ML 计划涉及整个组织的决策者
Trend 7: Maturity limited by alignment issues趋势 7：成熟度受限于对齐问题
Organizational alignment is the biggest gap in achieving AI/ML maturity
组织一致性是实现 AI/ML 成熟度的最大差距
The bottom line底线
Stuck in the lab Disconnected teams Technology mismatch Stakeholder buy-in Hidden technical debt Inefficient machine learning lifecycle |
困在实验室 断开连接的团队 技术不匹配 利益相关者的支持 隐藏的技术债务 低效的机器学习生命周期 |
Stuck in the lab: Compliance with existing IT governance, security, and auditability requirements delays or prevents deployment. Disconnected teams: Difficulty aligning data science development needs with IT requirements for production. Technology mismatch: Missed opportunities to deploy models in time to capitalize on market opportunities. Stakeholder buy-in: Difficulty tracking ML investment outcomes for value delivered. Hidden technical debt: Frequent updates, significant production testing, and constant validation required to maintain model quality and performance. |
困在实验室：遵守现有的 IT 治理、安全性和可审计性要求会延迟或阻止部署。 断开连接的团队：难以将数据科学开发需求与 IT 产品需求保持一致。 技术不匹配：错过了及时部署模型以利用市场机会的机会。 利益相关者的支持：难以跟踪 ML 投资结果以实现价值交付。 隐藏的技术债务：维护模型质量和性能所需的频繁更新、重要的产品测试和持续验证。 |
Technical debt is piling up技术债务堆积如山
Trend 8: Deployment time is increasing趋势 8：部署时间正在增加
Only 11% of organizations can put a model into production within a week,and 64% take a month or longer
只有 11% 的组织可以在一周内将模型投入生产，而 64% 的组织需要一个月或更长时间
The time required to deploy a model is increasing year-on-year
部署模型所需时间逐年增加
Trend 9: Data scientists spend too much time on deployment趋势 9：数据科学家在部署上花费过多时间
38% of organizations spend more than 50% of their data scientists' time on deployment
38% 的企业将超过 50% 的数据科学家时间用于部署
Organizations with more models spend more of their data scientists' timeon deployment, not less
拥有更多模型的企业将更多的数据科学家的时间花在部署上，而不是更少
MLOps preferences偏好
Trend 10: Improved outcomes with MLOps solutions趋势 10：使用 MLOps 解决方案改善结果
71% of all organizations have hybrid environments, and 42% have acombination of cloud and on-premises infrastructure
71% 的组织拥有混合环境，42% 拥有云和本地基础架构的组合
42% of respondents have hybrid environments consisting of both cloud and on-premises solutions.
Last year, only 16% did.
42% 的受访者拥有由云和本地解决方案组成的混合环境。
去年，只有 16% 的人这样做了。
Buying a third-party solution costs 19-21% less than building your own |
购买第三方解决方案的成本比构建自己的解决方案低 19-21% |
Respondents were asked to indicate theiraverage annual infrastructure costs based onpredefined ranges, such as "$51-$100K". Thetotal average annual infrastructure cost wasthen estimated as a range.The low estimate isbased on the lower bound for each predefinedrange (for example, $51K for “$51-$100K").The high estimate is based on the upper boundfor each predefined range (for example, $100Kfor"$51-$100K"). For the pre-defined rangethat represented the greatest cost ("more than$10M"), the lower bound of the range was usedfor both the high and low estimate.The percentdifference was calculated with the underlyingdata before rounding to the nearest percentage point. |
受访者被要求根据预定义的范围（例如“$51-$100K”）说明他们的平均年度基础设施成本。 然后将年平均基础设施总成本估计为一个范围。低估计基于每个预定义范围的下限（例如，$51K 对应“$51-$100K”）。高估计基于每个预定义范围的上限（ 例如，$100K 代表“$51-$100K”）。对于代表最大成本（“超过 1000 万美元”）的预定义范围，该范围的下限用于最高和最低估计。百分比差异为 在四舍五入到最接近的百分点之前用基础数据计算。 |
Organizations that buy a third-party solution spend less of theirdata scientists' time on model deployment
购买第三方解决方案的组织在模型部署上花费的数据科学家时间更少
The time required to deploy a model is 31% lower for organizations thatbuy a third-party solution
对于购买第三方解决方案的组织来说，部署一个模型所需的时间降低了31%
Conclusion
2021: The year of ML Next year will be a crucial year for AI/ML initiatives.There’s increased urgency—don’t get left behind. More accessible than ever Despite the increasing complexity of the space, it’s never been easier to start investing in AI/ML and scale it more effectively. You need MLOps Organizations that invest in operational efficiency will reap the greatest benefits in 2021. The time to act is now! |
2021：机器学习之年 明年将是 AI/ML 计划的关键一年。紧迫性越来越高——不要落后。 比以往任何时候都更容易访问 尽管该领域的复杂性越来越高，但开始投资 AI/ML 并更有效地扩展它从未如此简单。 你需要 MLOps 投资于运营效率的组织将在 2021 年获得最大的收益。现在是行动的时候了！ |
