Automated Machine Learning (Auto ML) may be the essential technology required by supply chain and industrial IoT users to generate value from their IoT and digital twin investments. Companies like IBM, DataRobot, Altryx, Google, AWS, Microsoft, Dataiku, H2O and numerous others are developing new low code and no code approaches to adding machine learning to our IoT deployments. IBM uses the term “AI for AI” to describe the capability to use existing AI resources to develop new AI solutions. Auto ML is a rapidly growing market with many participants, and new products coming to market on a regular basis. Featuring graphical drag and drop and wizard based low or no code interfaces these products are designed to be used by people that are not data scientists. A robust Auto ML tool will support users with the heavy lift of starting an ML project including
Data preparation (selection, cleaning, formatting)
Algorithm testing and selection
Model training and tuning (hyperparameter optimization)
The focus is on automating machine learning development for a broad range of companies that lack the skilled talent necessary to do this on the own with traditional, labor intensive approaches to data science.
For IoT users in industrial manufacturing and the supply chain this could open up their digital twins beyond the simple process or transaction data historian role that they are playing today.
To read more on the actual mechanics of the process take a look at this blog from IBM Research –
This includes links to some really interesting (and much deeper) papers on the topic of Auto ML presented at AAAI 2020.
Want to try this for yourself? Microsoft has a very simple tutorial to take the Azure Machine Learning Studio for a test drive –
While there are a number of limitations to the Auto ML approach, for many companies this represents the only viable way to get started in the near term. Once you get started invest some time in learning about ML Ops – think DevOps for ML. Models require management just like any other type of production code. The data science and the operations professionals must work together to manage the ongoing support and development of ML models and this code has a lifecycle that needs to be monitored and maintained.