AutoML – Accelerate, Democratize and Scale AI
The most fascinating thing about machine learning (ML) is that it can be applied to itself to accelerate and scale its own development, which is the core idea behind automatic machine learning (AutoML). Gartner predicts that by 2020, more than 40% of the data science tasks would have the possibility of being automated. AutoML aims to do just that, makes decisions in a data-driven, objective, and automatic way.
The user is supposed to just feed in data, and the AutoML system automatically determines the approach that performs the best for the particular problem and builds an ML model. It aims to automate the entire AI workflow. All the user needs to do is feed in data and validate the model. It can take care of data cleaning, feature pre-processing, model selection and parameter optimization. Some of the key AutoML platforms are by H2O.ai, TPOT, Amazon, Google and Microsoft
Can AutoML replace data scientists?
The short answer is no, AutoML cannot replace data scientists, in fact, it will aid data scientists so that they focus on human tasks and the repetitive tasks can be taken care of by AutoML.
By commoditizing ML for process improvement, it does pose a question on what the interplay between data, models, and human experts should look like. While AutoML is good at data pre-processing, parameter tuning and model building, they are still not capable of doing most of what a data scientist does. AutoML cannot define business problems, or apply domain knowledge to draw actionable insights from data.
Further, it empowers “citizen data scientists”, employees having valuable domain expertise, but not possessing a programming background, by helping them translate their knowledge into the AI world. It “democratizes AI” by allowing organisations with limited data science expertise to develop analytical capabilities to solve sophisticated business problems.
Key market players
Research by Forrester highlighted that solutions developed with advanced feature engineering capabilities and model transparency were key differentiators for Leaders in the AutoML space, empowering citizen data scientists and data scientists alike to tackle more-challenging use cases.
These AutoML tools automate the end-to-end life cycle of developing and deploying predictive model, starting from data preparation phase through feature engineering, model training, validation, and ModelOps.
Some key advantages of using AutoML
- Aids data scientists by automating repetitive, time-consuming tasks of a machine learning process
- Enables citizen data scientists to build accurate models with without actually programming
- Provides interpretability and explainability for all models – doesn’t act as a black box
- Provides a quick path for building production ready AI models
Want to know more on latest AI trends for this TECHADE, read our article Demystifying Tech for the TECHADE: Artificial Intelligence (AI)