Newsletter & AutoML!

First, welcome to my newsletter! I feel honored for the early access to LinkedIn's recent product, Newsletter! If you have been following my posts, please subscribe so that I feel committed to continue with my Sunday updates.

Second, I have selected a simple and relevant name for it; TRES. I would like to dedicate my notes to three topics I'm most passionate about: Technology, Business Strategy and Product. Each week, I will select one area and will share my thoughts with you.

Third, I would like to continue with Sunday releases. If you feel I need to change the cadence or add more topics, please feel free to connect with me directly!

Needless to say, all opinions expressed are solely my own and do not express the views of any entity.

Now - let's get back to our usual. In my last article, I discussed what AutoML is and why many companies are investing heavily in this space. In this article, I would like to introduce three products by three giants with an aim to automate ML:

1. AWS: Amazon Personalize & SageMaker Autopilot

Amazon personalize is an AWS service that enables personalized recommendations and targeted marketing on top of relevant applications (e.g., retail and gaming), aiming to maximize customer targeting and engagement. Use cases include user-based recommendations, similar-item recommendations, and personalized rankings (e.g., trending, seasonal, etc).

Referring to AutoML benefits on previous article, Amazon Personalize 1) shortens time-to-market for businesses looking to implement personalization in-house, 2) removes the need to machine learning and data science experts to train, test, and deploy personalization models in-house, and 3) reduces errors caused by human interactions and manual feature selection, and improves relevance due to incorporating large attribute sets.

How does it do it? It offers API-based services to simply feed the data (S3 or stream) into the system. At this point, Amazon Personalize will serve as a black box where it can automatically and without manual interactions "process and examine your data, identify what is meaningful, allow you to pick a machine learning algorithm, and train and optimize a custom model based on your data". Isn't this black-box automation just wonderful?! As an outcome, Amazon Personalize will generate personalized recommendations in real-time or in bulk. Below is the diagram obtained from its homepage. You can find some code samples here.

In addition to Amazon Personalize, AWS also offers SageMaker Autopilot which is aimed at automatically training and testing models on top of tabular data, with the aim to predict a column automatically. You simply feed your tabular data from S3 to Autopilot and select a column to be predicted. The Autopilot black-box will then do the magic. Here is a diagram provided on their homepage.

2. GCP: Cloud AutoML & AutoML Tables

Google Cloud Platform has also introduced Cloud AutoML in multiple data formats and use case, including Vision, Natural Language, and Structured data. Below is the diagram obtained from their homepage.

Among GCP's AutoML solutions, AutoML Tables seems similar to what SageMaker Autopilot is offering. You can feed your structured data into AutoML Tables from BigQuery or a CSV file, and define your data schema, target for prediction, and constraints (e.g., training time). AutoML Tables then automatically analyzes the data and searches its "model zoo" for the most relevant model from simple regressions to deep and ensemble methods. It trains the model and provides you with details of model evaluation to thoroughly check and select the best model. You can then deploy the selected model within the same solution, as shown below.

3. Microsoft Azure AutoML

Microsoft Azure also offers Automated ML; you feed your data, you define your target goals and constraints (e.g., training time). The AutoML then automatically "examines the data characteristics and recommends new pipelines to build your ML model. This includes preprocessing steps, feature generation, model selection, and hyper-parameter tuning.

Summary

Overall, all three companies offer similar experiences when it comes to AutoML. Their offerings significantly reduces time to train and test ML models, and will definitely enable software engineers and technical product managers to use these systems easily.

Among all, Amazon Personalize's specific solution for personalization has taken AutoML one step further. It is evident that the product is developed through multiple iterations and customer feedback to provide an end-to-end automation. Additionally, Amazon's deep personalization/recommendations knowledge and experience in eCommerce for the last 20 years has contributed to its success. This is a unique product that reminds us of the importance of vertical innovation in addition to horizontal automations.