W-JAX Munich 2017 - Agile Machine Learning: from Theory to Production

Artificial Intelligence (AI) and Machine Learning (ML) are all the rage right now. In this session, we’ll be looking at engineering best practices that can be applied to ML, how ML research can be integrated with an agile development cycle, and how open ended research can be managed within project planning According to a recent Narrative Science survey, 38 per cent of enterprises surveyed were already using AI, with 62 per cent expecting to be using it by 2018. So it’s understandable that many companies might be feeling the pressure to invest in an AI strategy, before fully understanding what they are aiming to achieve, let alone how it might fit into a traditional engineering team or how they might get it to a production setting. At Basement Crowd we are currently taking a new product to market and trying to go from a simple idea to a production ML system. Along the way we have had to integrate open ended academic research tasks with our existing agile development process and project planning, as well as working out how to deliver the ML system to a production setting in a repeatable, robust way, with all the considerations expected from a normal software project.

https://jax.de/en/

Transcript

1. From Theory to Production #TeamAwesome Agile Machine Learning
2. Who am I? CTO @ Basement Crowd rob_hinds robhinds
3. Agile Machine Learning Agile Machine Learning
4. Why should you care?
5. “Your AI will be a key point of distinction for your business” Accenture - Technology Visions 2017
6. “Products that don’t use [AI or ML] will die a natural death” Manish Singhal - Forbes India
7. 62% Percentage of organizations expecting to be using AI Technologies by 2018 Narrative Science - Outlook on Artificial Intelligence in the Enterprise 2016
8. https://spectrum.ieee.org/computing/software/the-2017-top-programming-languages
9. https://insights.stackoverflow.com/survey/2017
10. DON’T BELIEVE THE HYPE
11. “The first wave of corporate AI is doomed to fail” Harvard Business Review - The First Wave of Corporate AI Is Doomed to Fail
12. So, what can we do?
13. Sensible engineering & product design principles are the key
14. Product Thinking for Machine Learning
15. https://www.useronboard.com/features-vs-benefits/ Machine Learning != Your Product
16. Is Machine Learning part of your MVP?
17. Is your Machine Learning Mission Critical?
18. https://spectrum.ieee.org/automaton/robotics/artificial-intelligence/how-google-self-driving-car-works
19. Data (photo) Machine Learning (suggested tags) Curation (creator) Customers (consumers)
20. 3 Principles: 1) Don’t build Machine Learning for the sake of it 2) Do you need ML in your MVP to test product market fit? 3) Is your ML mission critical?
21. “a right to explanation” GDPR: Article 22
22. Engineering Thinking for Machine Learning
23. It’s still just Engineering
24. Clean code Testing Modularity
25. ML Anti-Patterns: Dead experiment code - Configuration debt Code glue - Pipeline jungles Sculley, D., et al. "Hidden technical debt in machine learning systems."
26. “Glue code and pipeline jungles are symptomatic of integration issues that may have a root cause in overly separated ‘research’ and ‘engineering’ roles” Sculley, D., et al. "Hidden technical debt in machine learning systems."
27. Agile Thinking for Machine Learning
28. Research sprints
29. Build > Measure > Learn Eric Ries - The Lean Startup
30. Simple Rule based Traditional off-the-shelf libraries Deep Learning & Sophisticated ML pipelines
31. Agile Machine Learning Agile Machine Learning
32. Text ➡ Numbers
33. Text ➡ Numbers AI pretends to fail Turing Test. 3 145 82 31 96 733 Bag-of-Words https://en.wikipedia.org/wiki/Bag-of-words_model
34. Text ➡ Numbers AI pretends to fail Turing Test. [1.25,...,3.58] [0.05,...,0.07] [45.8,...,9.70] [0.78,...,10.1] [100.1,...,7.8] [445.1,...,2.1] word2vec https://www.tensorflow.org/tutorials/word2vec
35. Demo
36. Theory to Production
38. Choosing your stack Available skills set Existing knowledge & tech stack Hiring pool
39. Modern architecture & cloud technology makes ML deployment easier
40. Photo by frank mckenna on Unsplash
41. https://pbs.twimg.com/media/C4vf8SQUcAALCyl.jpg
42. AWS nginx Zuul (Edge server) Eureka (service registry) Recommendation Service Movies Service
43. Take aways • Approach it with the rigour and principles of any other engineering product • De-risk the cost of failure with sensible product management • Engineer sensibly! • Use tried and tested build (CI) and deployment approaches
44. Thanks! (any questions?)
45. References 1. https://resources.narrativescience.com/Resources/Resource-Library/Article-Detail-Page/announcing -our-new-research-report-outlook-on-artificial-intelligence-in-the-enterprise-2016 2. https://www.accenture.com/us-en/insight-disruptive-technology-trends-2017 3. http://fortune.com/2016/06/03/tech-ceos-artificial-intelligence 4. https://hbr.org/2017/04/the-first-wave-of-corporate-ai-is-doomed-to-fail 5. http://theleanstartup.com/principles 6. http://martin.zinkevich.org/rules_of_ml/rules_of_ml.pdf 7. https://papers.nips.cc/paper/5656-hidden-technical-debt-in-machine-learning-systems.pdf 8. Photos from unsplash.com