AutoML

1. AutoML

SPEAKER: In the previous lesson, you learned about Vertex AI, a unified platform that supports both AutoML, a no-code solution, and custom training, a code-based solution. In this lesson, you'll explore AutoML in depth, including the technologies used to power automated ML development. AutoML, which stands for Automated Machine Learning, aims to automate the process to develop and deploy an ML model. If you've worked with AutoML models before, you know that building them can be extremely time consuming because you need to repeatedly add new data and features, try different models, and tune parameters to achieve the best result. When ML was first announced in January of 2018, the goal was to save the manual work from data scientists and automate machine learning pipelines from pre-processing data to model training and deployment. Since 2021, AutoML features are embedded in Vertex AI and have become part of the platform. But how could this be done? How can you trust AutoML to generate the best results without bias and fast? Let's look deeper to explore how AutoML works and the main technologies behind it. AutoML is powered by the latest research from Google. It's an ongoing endeavor. There are four distinct phases. Phase 1 is data processing. After you upload a data set, AutoML provides functions to automate part of the data preparation process. For example, it can convert numbers, datetime, text, categories, arrays of categories, and nested fields into a certain format of data so that it can be fed into an ML model. Phase 2 includes searching the best models and tuning the parameters. Two critical technologies support this auto search. The first one is called neural architecture search, which helps search the best models and tune the parameters automatically. And the second one is called transfer learning, which helps speed the search by using the pre-trained models. Let's first look at neural architecture search. The goal of neural architecture search is to find optimal models among many options. Specifically, AutoML tries different architectures and models and compares the performance of the models to find the best ones. For instance, AutoML can search through multiple advanced ML models and automatically tune the parameters to find the best fit for your data. Secondly, let's examine transfer learning. Machine learning is similar to human learning. It learns new things based on existing knowledge. AutoML has already trained many different models with large amounts of data. These trained models can be used as a foundation model to solve new problems with new data. A typical example are Large Language Models or LLMs, which are general purpose and can be pre-trained and fine-tuned for specific purposes. LLMs are trained for general purposes to solve common language problems, such as text classification, question answering, document summarization, and text generation across industries. The models can then be tailored to solve specific problems in different fields such as retail, finance, and entertainment using a relatively small size of field data sets. Transfer learning is a powerful technique that lets people with smaller data sets or less computational power achieve great results by using pre-trained models trained on similar larger data sets. Because the model learns through transfer learning, it doesn't have to learn from the beginning, so it can generally reach higher accuracy with much less data and computation time than models that don't use transfer learning. In phase 3, the best models are assembled from phase 2 and prepared for prediction in phase 4. Note that AutoML does not rely on one single model, but on the top number of models. The number of models depends on the training budget, but is typically around 10. The assembly can be as simple as averaging the predictions of the top number of models. Relying on multiple top models instead of one greatly improves the accuracy of prediction. By applying these advanced ML technologies, AutoML automates the pipeline from feature engineering to architecture search, to hyperparameter tuning, and to model ensemble. It might seem that AutoML can do a better job than a human to find the optimal models that fit your data. Perhaps the best feature of AutoML is that it provides a no-code solution. You can point and click through a UI to build an ML model with your own data. You'll walk through the details from preparing training data to training your model and finally get prediction in the next module.

2. Let's practice!

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.