More Related Content
Similar to Productizing Machine Learning at the Edge (20)
Productizing Machine Learning at the Edge
- 1. Productizing Learning at the Edge Vera Serdiukova in/veraserdiukova/
- 2. Agenda 1 Should You Train at the Edge 2 Making Big Decisions 3 Getting Your Hands Dirty
- 3. Definitions edge device – a device whose compute, memory, and energy resources are constrained and cannot be easily changed. learning at the edge – (re)training of a machine learning model on the edge device. Source: https://arxiv.org/pdf/1911.00623.pdf
- 8. Should You Train at the Edge
- 12. Should I train at the Edge?
- 13. Do you have a clear understanding of the target device constraints?
- 14. Do you have a clear understanding of the target device constraints? N
- 15. Don’t do it. N
- 16. Don’t do it. N … but I can pick my own hardware
- 17. N … but I can pick my own hardware Are you/your users willing to pay more for a potentially more expensive device?
- 18. N … but I can pick my own hardware Are you/your users willing to pay more for a potentially more expensive device? N
- 19. Don’t do it. N
- 20. N … but I can pick my own hardware Are you/your users willing to pay more for a potentially more expensive device? N Y
- 21. Do you have a clear understanding of the target device constraints? Y Y
- 22. Do you have a sufficient amount of qualifying training data on the device? Y Y
- 23. Do you have a sufficient amount of qualifying training data on the device? N
- 24. Can you obtain the data you need through the target device? N
- 25. Can you obtain the data you need through the target device? N N
- 26. Don’t do it. N
- 27. Can you obtain the data you need through the target device? N N Y
- 28. Y Y Do you have a sufficient amount of qualifying training data on the device? Y
- 29. Do the existing ML tools satisfy your on-device training objectives? Y Y
- 33. Do the existing ML tools satisfy your on-device training objectives? N
- 34. Do you have an existing expertise in Machine Learning and Embedded Engineering and/or willing to invest heavily? N
- 35. Do you have an existing expertise in Machine Learning and Embedded Engineering and/or willing to invest heavily? N N
- 36. Don’t do it. N
- 37. Do you have an existing expertise in Machine Learning and Embedded Engineering and/or willing to invest heavily? N N Y
- 38. Y Y Do the existing ML tools satisfy your on-device training objectives? Y
- 40. Congratulations! Sounds like Learning at the Edge might be an eligible candidate for your needs. Y Y
- 42. 1/ What do you want to learn through training at the Edge?
- 46. 2/ Where and how does the learning happen?
- 52. 3/ When does the learning happen?
- 54. 4/ What is the role of your user in the training process?
- 56. Getting Your Hands Dirty
- 57. Design to obtain data
- 59. Enable data labeling add a name
- 61. Build model validation mechanisms
- 62. Source: https://blog.google/products/search/gboard-now-on-android// alternatives “manual” control Design for course-correction
- 63. Should You Train at the Edge Making Big Decisions Getting Your Hands Dirty