MLOps

2020-03-16T08:28:47Z

42.200.145.40: /* Architecture */

'''MLOps''' (a compound of “[[machine learning]]” and “operations”) is a practice for collaboration and communication between [[data scientists]] and operations professionals to help manage production ML (or [[deep learning]]) lifecycle.<ref>{{cite web |last1=Talagala |first1=Nisha |title=Why MLOps (and not just ML) is your Business’ New Competitive Frontier |url=https://aitrends.com/machine-learning/mlops-not-just-ml-business-new-competitive-frontier/ |website=AITrends |publisher=AITrends |accessdate=30 January 2018}}</ref> Similar to the [[DevOps]] or [[DataOps]] approaches, MLOps looks to increase automation and improve the quality of production ML while also focusing on business and regulatory requirements. While MLOps also started as a set of best practices, it is slowly evolving into an independent approach to ML lifecycle management. MLOps applies to the entire lifecycle - from integrating with model generation ([[software development lifecycle]], [[continuous integration]]/[[continuous delivery]]), orchestration, and deployment, to health, diagnostics, governance, and business metrics.

== History ==

The challenges of the ongoing use of machine learning in applications were highlighted in a 2015 paper titled, Hidden Technical Debt in Machine Learning Systems.<ref>{{cite journal |last1=Sculley |first1=D. |last2=Holt |first2=Gary |last3=Golovin |first3=Daniel |last4=Davydov |first4=Eugene |last5=Phillips |first5=Todd |last6=Ebner |first6=Dietmar |last7=Chaudhary |first7=Vinay |last8=Young |first8=Michael |last9=Crespo |first9=Jean-Francois |last10=Dennison |first10=Dan |title=Hidden Technical Debt in Machine Learning Systems |journal=NIPS Proceedings |date=7 December 2015 |issue=2015 |url=https://papers.nips.cc/paper/5656-hidden-technical-debt-in-machine-learning-systems.pdf |accessdate=14 November 2017}}</ref>

The predicted growth in machine learning includes an estimated doubling of ML pilots and implementations from 2017 to 2018, and again from 2018 to 2020.<ref>{{cite web |last1=Sallomi |first1=Paul |last2=Lee |first2=Paul |title=Deloitte Technology, Media and Telecommunications Predictions 2018 |url=https://www2.deloitte.com/content/dam/Deloitte/global/Images/infographics/technologymediatelecommunications/gx-deloitte-tmt-2018-predictions-full-report.pdf |website=Deloitte |publisher=Deloitte |accessdate=13 October 2017}}</ref> Spending on machine learning is estimated to reach $57.6 billion by 2021, a compound annual growth rate (CAGR) of 50.1%.<ref>{{cite web |last1=Minonne |first1=Andrea |last2=Schubmel |first2=David |last3=George |first3=Jebin |last4=Piña |first4=Jeronimo |last5=Danqing Cai |first5=Jessie |last6=Leung |first6=Jonathan |last7=Dimitrov |first7=Lubomir |last8=Ranjan |first8=Manish |last9=Daquila |first9=Marianne |last10=Kumar |first10=Megha |last11=Iwamoto |first11=Naoko |last12=Anand |first12=Nikhil |last13=Carnelley |first13=Philip |last14=Membrila |first14=Roberto |last15=Chaturvedi |first15=Swati |last16=Manabe |first16=Takashi |last17=Vavra |first17=Thomas |last18=Zhang |first18=Xiao-Fei |last19=Zhong |first19=Zhenshan |title=Worldwide Semiannual Artificial Intelligence Systems Spending Guide |url=https://www.idc.com/getdoc.jsp?containerId=IDC_P33198 |publisher=IDC |accessdate=25 September 2017}}</ref>

Reports show a majority (up to 88%) of corporate AI initiatives are struggling to move beyond test stages. However, those organizations that actually put AI and machine learning into production saw a 3-15% profit margin increases.<ref>{{cite web |last1=Bughin |first1=Jacques |last2=Hazan |first2=Eric |last3=Ramaswamy |first3=Sree |last4=Chui |first4=Michael |last5=Allas |first5=Tera |last6=Dahlström |first6=Peter |last7=Henke |first7=Nicolaus |last8=Trench |first8=Monica |title=Artificial Intelligence The Next Digital Frontier? |url=https://www.mckinsey.com/~/media/McKinsey/Industries/Advanced%20Electronics/Our%20Insights/How%20artificial%20intelligence%20can%20deliver%20real%20value%20to%20companies/MGI-Artificial-Intelligence-Discussion-paper.ashx |website=McKinsey |publisher=McKinsey Global Institute |accessdate=1 June 2017}}</ref>

In 2018, MLOps and approaches to it began to gain traction among AI/ML experts, companies, and technology journalists as a solution that can address the complexity and growth of machine learning in businesses.<ref>{{cite web |last1=G |first1=Doug |title=MLOps Silicon Valley |url=https://www.meetup.com/MLOps-Silicon-Valley/?_cookie-check=o1SkbKRfUlSuQoT3 |website=Meetup |publisher=Meetup |accessdate=2 February 2018}}</ref> <ref>{{cite web |last1=Bridgwater |first1=Adrian |title=Should every business function have an Ops extension? |url=https://techhq.com/2018/04/should-every-business-function-have-an-ops-extension/ |website=Tech HQ |publisher=Tech HQ |accessdate=13 April 2018}}</ref> <ref>{{cite web |last1=Royyuru |first1=Avinash |title=How to build AI culture: go through the curve of enlightenment |url=https://hackernoon.com/how-to-build-ai-culture-go-through-the-curve-of-enlightenment-21c239c1d5a7 |website=Medium |publisher=Hackernoon |accessdate=28 April 2018}}</ref> <ref>{{cite web |last1=Talagala |first1=Nisha |title=Why MLOps (and not just ML) is your Business’ New Competitive Frontier |url=https://aitrends.com/machine-learning/mlops-not-just-ml-business-new-competitive-frontier/ |website=AITrends |publisher=AITrends |accessdate=30 January 2018}}</ref> <ref>{{cite web |last1=Simon |first1=Julien |title=MLOps with serverless architectures (October 2018) |url=https://www.slideshare.net/JulienSIMON5/mlops-with-serverless-architectures-october-2018 |website=LinkedIn SlideShare |publisher=Julien Simon |accessdate=23 October 2018}}</ref> <ref>{{cite web |last1=Saucedo |first1=Alejandro |title=Scalable Data Science/Machine Learning: The State of DataOps / MLOps in 2018 |url=http://www.machinelearning.ai/machine-learning/alejandro-saucedo-scalable-data-sciencemachine-learning-the-state-of-dataops-mlops-in-2018/ |website=MachineLearning.AI |publisher=Alejandro Saucedo |accessdate=9 September 2018}}</ref> <ref>{{cite web |last1=Nicholson |first1=Chris |title=AI Infrastructure & Machine Learning Operations (AIOps or MlOps) |url=https://blog.skymind.ai/ai-infrastructure-machine-learning-operations-mlops/ |website=Skymind Blog |publisher=Skymind |accessdate=13 June 2018}}</ref> <ref>{{cite web |last1=Talagala |first1=Nisha |title=Operational Machine Learning: Seven Considerations for Successful MLOps |url=https://www.kdnuggets.com/2018/04/operational-machine-learning-successful-mlops.html |website=KDNuggets |publisher=KDNuggets |accessdate=1 April 2018}}</ref> <ref>{{cite web |last1=Banks |first1=Erink |title=BD Podcast Ep 34 – Putting AI to Work with MLOps Powered by ParallelM |url=https://bigdatabeard.com/bd-podcast-ep-34-putting-ai-to-work-with-mlops-powered-by-parallelm/ |website=Big Data Beard |publisher=Big Data Beard |accessdate=17 July 2018}}</ref> <ref>{{cite web |last1=Sato |first1=Kaz |title=What is ML Ops? Solutions and best practices for applying DevOps to production ML services |url=https://conferences.oreilly.com/artificial-intelligence/ai-eu-2018/public/schedule/detail/68247 |website=Artificial Intelligence Conference |publisher=O'Reilly |accessdate=10 October 2018}}</ref>

== Architecture ==

There are a number of barriers that prevent organizations from successfully implementing ML across the enterprise, including difficulties with:<ref>{{cite web |last1=Walsh |first1=Nick |title=The Rise of Quant-Oriented Devs & The Need for Standardized MLOps |url=http://slides.com/walsh/standards-in-ml-ops#/ |website=Slides |publisher=Nick Walsh |accessdate=1 January 2018}}</ref>
* Deployment and automation
* Reproducibility of models and predictions<ref>{{cite web |last1=Warden |first1=Pete |title=The Machine Learning Reproducibility Crisis |url=https://petewarden.com/2018/03/19/the-machine-learning-reproducibility-crisis/ |website=Pete Warden's Blog |publisher=Pete Warden |accessdate=19 March 2018}}</ref>
* Diagnostics<ref>{{cite web |last1=Warden |first1=Pete |title=The Machine Learning Reproducibility Crisis |url=https://petewarden.com/2018/03/19/the-machine-learning-reproducibility-crisis/ |website=Pete Warden's Blog |publisher=Pete Warden |accessdate=10 March 2018}}</ref>
* Governance and regulatory compliance<ref>{{cite web |last1=Vaughan |first1=Jack |title=Machine learning algorithms meet data governance |url=https://searchdatamanagement.techtarget.com/feature/Machine-learning-algorithms-meet-data-governance |website=SearchDataManagement |publisher=TechTarget |accessdate=1 September 2017}}</ref>
* Scalability<ref>{{cite web |last1=Lorica |first1=Ben |title=How to train and deploy deep learning at scale |url=https://www.oreilly.com/ideas/how-to-train-and-deploy-deep-learning-at-scale |website=O'Reilly |publisher=O'Reilly |accessdate=15 March 2018}}</ref>
* Collaboration<ref>{{cite web |last1=Garda |first1=Natalie |title=IoT and Machine Learning: Why Collaboration is Key |url=https://www.iottechexpo.com/2017/10/ai/iot-machine-learning-ml-ai-why-collaboration-key/ |website=IoT Tech Expo |publisher=Encore Media Group |accessdate=12 October 2017}}</ref>
* Business uses<ref>{{cite web |last1=Manyika |first1=James |title=What’s now and next in analytics, AI, and automation |url=https://www.mckinsey.com/featured-insights/digital-disruption/whats-now-and-next-in-analytics-ai-and-automation |website=McKinsey |publisher=McKinsey Global Institute |accessdate=1 May 2017}}</ref>

A standard practice, such as MLOps, takes into account each of the aforementioned areas, which can help enterprises optimize workflows and avoid issues during implementation.

A common architecture of an MLOps system would include data science platforms where models are constructed and the analytical engines were computations are performed, with the MLOps tool orchestrating the movement of machine learning models, data and outcomes between the systems.<ref>{{cite web |last1=Walsh |first1=Nick |title=The Rise of Quant-Oriented Devs & The Need for Standardized MLOps |url=http://slides.com/walsh/standards-in-ml-ops#/ |website=Slides |publisher=Nick Walsh |accessdate=1 January 2018}}</ref>

== See also ==
* [[IT_operations_analytics#AIOps|AIOps]]

== References ==
<references />

[[Category:Deep learning]]

Wikipedia - Benutzerbeiträge [de]

MLOps