The 12 Steps to Machine Learning and AI

let's do a quick review of what we've learned the 12 steps always start at step zero figure out where the machine learning is actually what you need and find a good application for it and so there you're thinking about what labels or outputs you actually want from your system then in step one figure out what it means to do good behavior here what are your goals which mistakes are worse than which other mistakes step 2 get your data step 3 split that data step four look at some of your training data step five get some tools and then step six actually train your models the step six is finished when you've got your candidate models and you know how they performed in your training data and you expect that their performance in training is better than their performance elsewhere and so if they're not meeting your bar in training don't proceed keep keep on keeping on in here then tune and debug that step is finished when you have used a separate data set which you've taken out of training if you want to get debugging opportunities and tuned hyper parameters next avoid bios remorse with validation so step 8 is finished when you have your best candidate model and you've evaluated success in fresh data and whether you want to go back and start again in the training process step nine testing should we actually allow this thing to move to the next stage get productionized step nine is finished when you have a document clearly detailing your decision process whether to take your model live and your estimated performance in your new data step 10 is productionizing a machine learning system that step is finished when you have your production ready system with automated retraining and safety nets now you make your launch decision in step 11. and then once step 11 is finished you've ramped up you've launched it now you're serving your model to users you're using a policy layer you're staying cautious step 12 that technical debt does your model stand the test of time so step 12 is never finished this is the gift that keeps giving but a good start is having a monitoring plan a maintenance plan tracking dashboards and stellar documentation alright so that's it for the course thanks so much for listening i hope you found these that 12 steps useful and enlightening [Applause] you

Channel: Cassie Kozyrkov

Views: 13,563

Keywords: DataScience, Data, DecisionIntelligence, MachineLearning, Statistics, AI, Analytics, Google, GoogleCloud, Education, ArtificialIntelligence, Decisions, Leadership, Technology, Cloud, Cassie, Cassie Kozyrkov, Tech, Google Cloud

Id: C_Q_L0wdPNg

Length: 3min 19sec (199 seconds)

Published: Mon Dec 19 2022