Automated Machine Learning  — The Next Big Thing

Original article was published by Aditya Ayyagari on Artificial Intelligence on Medium


AutoML — Adding Value to an Organization

The world of AutoML has been burgeoning over the past few years — and amid a looming economic recession, the idea of automating the development of AI and Machine Learning is bound to become even more appealing. We have talked about the typical features of AutoML. Now, let’s have a look at how it adds value to an organization:

Increased Automation

First-generation AutoML focused on automating the machine learning part of the data science process. Traditional data science workflow, however, includes a highly manual step called feature engineering which is the longest and most challenging part.

Platforms with “Automated Feature Engineering” capabilities now allow for the automated creation of features from relational data sources. This ability to “auto-generate” features in the data science process is a game-changing capability.

Increased Productivity

Creating models is the core activity for machine learning. It involves activities such as feature selection and engineering, data preparation, selection of algorithms, and evaluation and comparison of results. One of the most common benefits of using AutoML is increased productivity and effectiveness in creating analytical models and evaluating them.

Will AutoML replace Data Scientists??

Certainly not! The sole purpose of AutoML is to enhance the data scientist’s productivity. Automation can only help accelerate the process and make the field more open to other departments in an organization. The use of AI based features allows data scientists to explore the ‘unknown unknowns’ which data scientists generally don’t delve into.

Bridging the Skill Gap

It can help alleviate the skill shortage. Tasks like Feature selection, Hyperparameter tuning and choosing the Best Model — which demand a lot of the expertise — are automated.

Model and Feature Explanation

Machine learning models involve many different features or variables, and their relative importance in prediction or classification can be difficult to interpret.

In some industries such as financial services, model transparency and explainability (for example, for why a customer is extended or denied credit) are required by regulators. In the European Union, under the General Data Protection Regulations (GDPR), any citizen affected by an analytical model is guaranteed the “right to an explanation.” AutoML promises to provide this explanation.

Limitations

Data labeling and preparation

AutoML automates the cycle assuming that there is data available. The task of labeling and preparing the data remains solitary.

Quality of data

The AutoML platforms identify the best features and models based on the given data, irrespective of its quality. We only get quantitative results on the data.

Task specific

Majority of AutoML platforms are designed only for most common use-cases.

Explainability

AutoML platforms fail to add CONTEXT to the data, operations or their explanation.

Popular AutoML platforms

In my opinion

It seems inevitable that machine learning and other analytical approaches will be increasingly automated over time. Just like in other areas of business and life, approaches and processes that employ greater levels of intelligence and automation will eventually prevail. With a powerful tool that can bridge the skills gap and boost the performance of data scientists, organizations that can adopt this technology are likely to be much more productive.