Understand the ongoing cost and success criteria as part of your ML strategy
Start working on your machine learning inventory. You need to know where and why the organization has applied machine learning. Chances are it is built into some things that have been deployed along the way. Seriously you need to get that list built out. That is where it all starts to come together as a strategy. A recent survey of 1,870 organizations found that machine learning strategies in the wild have a long way to go for 80% of respondents.[1] You can read the press release about that survey/study and download it if you fill out a form.[2]
Where are you using machine learning and artificial intelligence in the organization?
Is it part of a budget driven machine learning strategy with actively monitored key performance indicators and proven return on investment?
What part of the budget is allocated for machine learning?
What is your machine learning key performance indicator?
What is your machine learning strategy?
Has your return on investment model been tested in practice?
To really understand the ongoing cost and success criteria of your ML strategy you need a pulse of what is going on and what will be going on in the future. We all have limited time and energy to put into figuring things out. This one of those things that you want to make sure you get right from the start. Your machine learning and artificial intelligence strategy has so much potential upside that getting it wrong could reverberate for years. Sometimes being early to the party makes doing the hard things nearly impossible. Right now machine learning in the enterprise is moving beyond the early party stage into the rationalized and empirically considered phase of planning and deployment.
Some of the actual training costs that people have advertised online are very large.[3] Anybody that was using a cloud system for training would probably have a reaction to a cost of several hundred thousands dollars. My general training costs have always been tagged to either electricity from my personal computer system or within other environments where those costs did not come directly back to my credit card. I started reading a post series from Brad Cordova on Medium about machine learning surprises.[4] It even had a reference to a paper on hidden technical debt in machine learning which brought me great joy on this cold Sunday morning.[5]
Go check out @robmay's new podcast over at @buzzsprout "Investing In AI Episode1: Rob Toews From Highland Capital" at https://buzzsprout.com/1697077/7888999
Fun Twitter hashtags #MachineLearning #ML #ArtificialIntelligence #AI #MLOps
If you really enjoy podcasts and missed my podcast recommendation from last week, then go check out this podcast on MLOps called “Delivery, Interrupted” https://anchor.fm/delivery-interrupted/episodes/Introduction-to-MLOps---Part-1-eqmob5
Or part 2 here:
https://anchor.fm/delivery-interrupted/episodes/MLOps-Part-2---Continuous-Delivery-ercgcv
Footnotes:
[1] https://venturebeat.com/2021/02/25/why-machine-learning-strategies-fail/
[3] You can see the cost from an Elliot Turner tweet here
or read more about that here https://medium.com/syncedreview/the-staggering-cost-of-training-sota-ai-models-e329e80fa82 or you can get to the actual paper here https://arxiv.org/abs/1906.08237
[4] https://medium.com/mysuperai/7-costly-surprises-of-machine-learning-part-seven-703f4a0298dd
[5] This paper was linked in footnote [4] and for some reason I really enjoyed it https://papers.nips.cc/paper/2015/file/86df7dcfd896fcaf2674f757a2463eba-Paper.pdf
What’s next for The Lindahl Letter?
Week 7: Plan to grow based on successful ROI
Week 8: Is the ML we need everywhere now?
Week 9: Valuing ML use cases based on scale
Week 10: Model extensibility for few shot GPT-2
Week 11: What is ML scale? The where and the when of ML usage
I’ll try to keep the what’s next list forward looking with at least five weeks of posts in planning or review. If you enjoyed reading this content, then please take a moment and share it with a friend.