Automated Machine Learning: An Introduction

What is Automated Machine Learning?

Automated Machine Learning (AutoML) is a field of machine learning concerned with automating the selection and implementation of various machine learning techniques. AutoML draws on a variety of machine learning disciplines, such as Bayesian optimization, various regression models, meta learning, transfer learning and combinatorial optimization.

There are typically three broad components of an AutoML platform:

1. Data Cleaning –  This step includes removing outliers, accounting for missing data, and merging data from various sources
2. Feature Engineering – This step focuses on creating a super-set of features from the raw inputs to the system. A typical example: it may be more useful to look at a user’s change in behavior over time. This is because it is often a derivative of their behavior. And this often supersedes their raw activity when you want to build a predictive churn algorithm. Feature engineering is a critical component of successful machine learning algorithm solutions. The set of possible super features is limitless. As a result, this step is often one of the most difficult to automate
3. Model Selection – The final step in an AutoML platform is optimizing the appropriate machine learning model. There are literally thousands of possible machine learning models to choose from ranging from deep learners to regression models to boosted decision trees. It’s very difficult to ascertain, a priori, which model will perform best on a set of data thus necessitating trying numerous models to obtain best performance. Implicit in model selection is a multi-step process. This also includes breaking up the input data into training, testing, and validation sets, and feature selection

Note that each of these steps, when AutoML is not present, can necessitate a significant investment of resources and time to implement correctly. Implementing AutoML platforms can significantly decrease the time to market for companies deploying ML technology.

Why use AutoML?

AutoML has several different prominent use cases within organizations including:

1. Enabling the fast deployment of ML solutions across an organization
2. Enabling data science teams to focus on specific portions of the AutoML pipeline. For instance, you can create new models or objective functions. At the same time, you can leave some of the other areas to the AutoML platform
3. Empowering teams to quickly benchmark and validate the performance of their own internal ML solutions

Cortex Queries – AutoML for Everyone in the Company

Vidora’s AutoML platform, Cortex, provides an additional capability through Cortex Queries which enables organizations to easily access the output of the AutoML engine by asking questions. In effect, Cortex Queries enables everyone from the C-suite to product managers to marketers, the ability to use machine learning technology directly.

The Future of AutoML

AutoML is a relatively new area of exploration and we expect rapid progress over the next few years. And over time, these advancements will result in an ever expanding set of problems being well suited for AutoML. Products like Cortex are at the forefront of AutoML and enabling Fortune 500 companies to leverage the technology today to automate and optimize their business using ML.