O slideshow foi denunciado.
Utilizamos seu perfil e dados de atividades no LinkedIn para personalizar e exibir anúncios mais relevantes. Altere suas preferências de anúncios quando desejar.
Próximos SlideShares
What to Upload to SlideShare
Avançar
Transfira para ler offline e ver em ecrã inteiro.

Compartilhar

"ML Services - How do you begin and when do you start scaling?" - Madhura Dudhgaonkar, Workday

Baixar para ler offline

Dynamic talks SF: So you have heard all the hype around how Machine Learning is going to change the world. But within your business context, where do you start? How do you get leadership buy-in for investment? And how and when you start scaling your ML Services?

In this session, you will walk away with an actionable framework to bootstrap and scale a machine learning services team. You will see the framework in action through an actual 0 to 1 product journey involving deep learning where we delivered value in record speed in-spite of not having a dataset when we started. You will get practical tips on how to make decisions about when and how to scale your capability to scale ML Services and platform. Some of the tips are pretty counterintuitive and revealed themselves with our experience of productizing ML services over the last 5+ years. (Using a diverse range of technologies - Vision, Language, Graph, Anomaly Detection, Search Relevance, Personalization)

  • Seja a primeira pessoa a gostar disto

"ML Services - How do you begin and when do you start scaling?" - Madhura Dudhgaonkar, Workday

  1. 1. ML Services - How to Begin Well and Scale Right Madhura Dudhgaonkar, Senior Director - Machine Learning
  2. 2. This presentation may contain forward-looking statements for which there are risks, uncertainties, and assumptions. If the risks materialize or assumptions prove incorrect, Workday’s business results and directions could differ materially from results implied by the forward-looking statements. Forward-looking statements include any statements regarding strategies or plans for future operations; any statements concerning new features, enhancements or upgrades to our existing applications or plans for future applications; and any statements of belief. Further information on risks that could affect Workday’s results is included in our filings with the Securities and Exchange Commission which are available on the Workday investor relations webpage: www.workday.com/company/investor_relations.php Workday assumes no obligation for and does not intend to update any forward-looking statements. Any unreleased services, features, functionality or enhancements referenced in any Workday document, roadmap, blog, our website, press release or public statement that are not currently available are subject to change at Workday’s discretion and may not be delivered as planned or at all. Customers who purchase Workday, Inc. services should make their purchase decisions upon services, features, and functions that are currently available. Safe Harbor Statement
  3. 3. Agenda Workday Confidential Introduction A Practical Example of Machine Learning 0 to 1 Product Journey Framework to Begin Well and Scale Right
  4. 4. Workday Inc. Madhura Dudhgaonkar Senior Director, Machine Learning
  5. 5. Mar 2009 - March 2006
  6. 6. Mar 2009 -March 2009 - Outrageous Thought Mt. Denali | 20,320 ft | 6,194 m
  7. 7. Leading provider of enterprise cloud applications Plan Execute Analyze Workday Planning Workday Financial Management Workday Human Capital Management Workday Prism Analytics and Benchmarking
  8. 8. Business Process Framework Object Data Model Reporting and Analytics Security Integration Cloud One Source for Data | One Security Model | One Experience | One Community Machine Learning One PlatformPower of One
  9. 9. Unique Data Sets 40%+ of Fortune 500 50%+ of Fortune 50 Clean operational data Critical business workflows
  10. 10. Workday Machine Learning Teams Workday Confidential Victoria, Canada Portland, OR Dublin, Ireland Boulder, CO San Francisco Pleasanton
  11. 11. Example 0 - 1 ML Product Journey Workday Confidential
  12. 12. A ML Service to scan receipts & auto-populate expense reports... The Challenge Workday Confidential
  13. 13. ...in Six Months! March Su Mo Tu We Th Fr Sa 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 April Su Mo Tu We Th Fr Sa 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 May Su Mo Tu We Th Fr Sa 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 June Su Mo Tu We Th Fr Sa 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 July Su Mo Tu We Th Fr Sa 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 August Su Mo Tu We Th Fr Sa 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 Workday Confidential
  14. 14. A ML Service that scans receipts with >80% accuracy that is delivered in private and public cloud for all customers by end of September. Define the Win Workday Confidential
  15. 15. Optical Character Recognition - Why build? Workday Confidential
  16. 16. First obstacle - data bootstrap Workday Confidential Deep Commitment to Ethics. Trust and Transparency about data with customers since day one.
  17. 17. Transferred Learning Data Workday Confidential
  18. 18. Variety of Text Workday Confidential
  19. 19. Step 1 - Bounding Box Detection A deep learning model based on Residual Networks Output: center of the box, height, width and angle of tilt, confidence Workday Confidential
  20. 20. Step 2 - Text Recognition A deep learning model based on Residual Networks Output: text Workday Confidential TOTAL 63.87 Brooks Brother 01/27/15
  21. 21. Step 3 - Mapping An assortment of models (rule based + deep learning/ensembles) Output: a map of value to field Workday Confidential 63.87 Brooks Brother 01/27/15 Total Merchant Date Shipping N/A
  22. 22. Production: Ensemble Workday Confidential
  23. 23. Human In The Loop Minimum Viable Product Data Cleansing Data Transformation Data Labelling Feature Engineering Model Selection Training Validation: Results and UX Metrics Selection Model & UX Exploration Validation Data cleansing and labelling Validation: Results and User Experience Workday Confidential
  24. 24. Victory! Productized >90% accuracy
  25. 25. Framework to Begin Well and Scale Right Workday Confidential
  26. 26. For 0 to 1 - START! Workday Confidential
  27. 27. S - Select One Win (Unambiguous Value) Workday Confidential Be Precise!
  28. 28. T - Team (<1 pie, 3-6 months) Workday Confidential
  29. 29. A - Articulate and Align Workday Confidential Articulate (Win & Gameplan) Align (Stakeholders)
  30. 30. R - Rally & Support
  31. 31. T - Take Shortcuts (tech debt - accrue it!) Workday Confidential
  32. 32. For 0 to 1 - START! Workday Confidential S - Select the first win T - Team (<1 pie, 3-6 months) A - Articulate (win/game plan) Align (stakeholders) R - Rally & Support T - Take shortcuts
  33. 33. To Scale - GET! Workday Confidential
  34. 34. G - Get Credit and Gather Capital Workday Confidential First win in the bag!
  35. 35. E - Establish Repeatable Processes & Platform Workday Confidential
  36. 36. T - Transfer and Apply Learnings from the 0 - 1 Workday Confidential
  37. 37. For Scale - GET! Workday Confidential G - Get credit for the first win, Gather more capital E - Establish repeatable processes and platform T - Transfer learnings to scale to 10 G
  38. 38. START & go GET ’em Workday Confidential
  39. 39. Hiring - Dev Managers, Engineers, PMs, Data scientists.

Dynamic talks SF: So you have heard all the hype around how Machine Learning is going to change the world. But within your business context, where do you start? How do you get leadership buy-in for investment? And how and when you start scaling your ML Services? In this session, you will walk away with an actionable framework to bootstrap and scale a machine learning services team. You will see the framework in action through an actual 0 to 1 product journey involving deep learning where we delivered value in record speed in-spite of not having a dataset when we started. You will get practical tips on how to make decisions about when and how to scale your capability to scale ML Services and platform. Some of the tips are pretty counterintuitive and revealed themselves with our experience of productizing ML services over the last 5+ years. (Using a diverse range of technologies - Vision, Language, Graph, Anomaly Detection, Search Relevance, Personalization)

Vistos

Vistos totais

320

No Slideshare

0

De incorporações

0

Número de incorporações

0

Ações

Baixados

4

Compartilhados

0

Comentários

0

Curtir

0

×