Why is your model stuck in the lab? How to move your model from the lab to production
0 likes24 views
The document outlines an agenda for a meeting led by Jeff Moore, covering key challenges in machine learning, such as unclear problem statements and integrating models into processes. It emphasizes the importance of effective communication, alignment with corporate goals, and leveraging data for predictive analytics. Additionally, it discusses strategies for building credibility in new data science initiatives within an organization.
Why is your model stuck in the lab? How to move your model from the lab to production
- 2. 1 AGENDA Agenda Item Topics Presenter(s) Time Allotment Welcome • Introduction • Jeff Moore • 5 mins Top ML Challenges • Data • Culture • Problem Statement / ROI • Tech • Change Management • 15 mins Mitigations • Marketing • Agile • OCM • Strategic Roadmap • 5 mins Use Case • Problem Statement • Plan of Attack • Results • 5 mins QA • Summary • 5 mins
- 4. 3 − Data Scientist – AI/ML Lead within Data and Process Transformation − Former Management Consultant − Former IT executive − Graduate School – Northwestern, NYU Stern − SCE email: jeff.moore@sce.com − Linkedin: linkedin.com/in/moorebigdata INTRODUCTION
- 5. 4 − Sources • https://medium.com/datadriveninvestor/data-science-challenges-b7622b85b807 • https://www.statista.com/statistics/1111249/machine-learning-challenges/ TOP ML CHALLENGES (2018 VS. 2020)
- 6. 5 − Lack of clear problem statement − Integrating the model into critical processes − Results used by decision makers − Meeting expectations and aligning the organization − Talk with stakeholders − Perform basic EDA, data visualizations − Formulate alternative problem statements − Embed the model into legacy applications − Make the model callable via API − Create meaningful metrics − Educate management on metric importance − Integrate metric into current reporting − Leverage change management − Link model to corporate goals or strategies − Educate, Communicate POTENTIAL MITIGATIONS
- 7. 6 − Problem Statement • When will the next injury event occur? − Data • We have all the data in the world − Advanced analytics is cool! Everyone will use it just because! − Everyone wants a new iphone! − Reframe Problem Statement • Can a risk framework be built that identifies safety patterns where risk is elevated? − Data – Actually we don’t! • EDA is your friend. Use it to show where there is plenty of data, but also where the gaps are. − Find suitable use cases for ML/AI • Educate, Market, Communicate • Be open to feedback from SMEs • Build model credibility − New iphone? No thanks! • Create value by integrating the model into existing workflow • Create new digital solutions to gain more, more frequent and accurate data USE CASE Context: Company that is new to predictive analytics. New data science community of practice. Numerous legacy systems and databases. Several, independent data science teams.
- 8. 7 SEEING ALIGNMENT IN PRACTICE… Assess crew and work risk – day of work Assess safety risk prior to executing work • Integrating model in other areas of interest: Public safety Data capture initiatives Metric reporting Work method training
- 9. 8 − Moving beyond the lab is more than just a model − Data is important but don’t forget to communicate, market and build alignment − A good model will have legs beyond the initial use case − Starting with a good problem statement can increase the likelihood of moving to production − Look to link your model to a strategic problem or corporate goal IN SUMMARY…
- 10. THANKS 9