AutoML will democratize machine learning

Like calculator has changed the way we do a simple calculation, #AutoML will democratize machine learning by making it simple for engineers and staff to use and eventually for the general people. As #IoT and #mobileML become mainstream, a new wave of AutoML tool kits would emerge enabling artificial intelligence capabilities in home electronics and gadgets. Things able to perform face recognition, adjust to environmental conditions, home demographics, daily/weekly uses trend with the flexibility of human interference to perform multiple machine learning operations and select one or two that best fit into current need.  

 Until recently I was of the view that AutoML toolsets are used for feature engineering but after doing a bit of research it is evident that AutoML has matured into the entire data science life cycle. This Forbes article https://www.forbes.com/sites/tomdavenport/2019/09/03/dotdata-and-the-explosion-of-automated-machine-learning/#1d9d84f82c3a (Links to an external site.) by Tom Davenport discusses some of the AutoML tools and how they cater to different personas.

A data science project is a complex endeavor as it involves multiple steps and requires different skill sets to successfully execute it. It typically starts with understanding the business domain, followed by understanding the data to create a data pipeline and doing feature engineering. The next step is model development and validation. After this, a selected model is productized.

Our ability to execute these projects is limited by the number of data scientists. We can’t analyze data; create enough good models with limited data scientists. AutoML powerful sets of tools help un all aspects of the life cycle including data analysis, feature engineering, generating well-fitting models using the best algorithms and finally generating APIs to deploy the model into production.

Organizations have been struggling with the speed to innovate using data that we have acquired into the data lake. Most have tried to democratize it by making it available to all via the workbench and Jupyter notebook but still, most of the data analysis and modeling work is limited to expert data scientists which have become the bottleneck. I believe that AutoML can at one hand expedite data scientists tasks allowing them more time and bandwidth to take additional talks but on the other hand enable a wide range of employee bases to explore data and perform tasks like feature engineering. 

A few AutoML tools to watch out for are: H20.ai; SageMaker Autopilot, Google AutoML, Azure AutoML, Auto-Keras, Auto Pytorch, and more.