Feels a bit overstated, but an interesting read on AutoML and its potential impacts on Data Science (and scientists)
Thereâ€™s a good reason for all the AutoML hype: AutoML is aÂ must-haveÂ for many organizations.
Letâ€™s take the example of Salesforce. TheyÂ explainÂ that their â€œcustomers are looking to predict a host of outcomes â€” from customer churn, sales forecasts and lead conversions to email marketing click throughs, website purchases, offer acceptances, equipment failures, late payments, and much more.â€
In short, ML is ubiquitous. However, for ML to be effective for each unique customer, they would â€œhave to build and deploy thousands of personalized machine learning models trained on each individual customerâ€™s data for every single use caseâ€ and â€œthe only way to achieve this without hiring an army of data scientists is through automation.â€
While many people see AutoML as a way to bring ease-of-use and efficiency to ML, the reality is that for many enterprise applications, thereâ€™s just no other way to do it. A company like Facebook or Salesforce or Google canâ€™t hire data scientists to build custom models for each of their billions of users, so they automate ML instead, enabling unique models at scale.
The amount of ML components that are automated depends on the platform, but with Salesforce, it includes feature inference, automated feature engineering, automated feature validation, automated model selection, and hyperparameter optimization.
Thatâ€™s a mouthful.
What this means is that data scientists can deploy thousands of models in production, with far less grunt work and hand-tuning, reducing turn-around-time drastically.
By shifting the work from data crunching towards more meaningful analytics, AutoML enables more creative, business-focused applications of data science.