Machine Learning and Artificial Intelligence on Google Cloud

The topics of machine learning (ML) and artificial intelligence (AI) are supercharged with societal imagination given the generative arms race that has kicked off between major companies such as Google, Amazon, and Microsoft, along with an ecosystem of model and solution providers such as Anthropic, Cohere, and Copy.ai. In this chapter, you will learn about the history of the usage of ML and AI at Google before we explore what solutions are available through Google Cloud.

By the end of this chapter, you will be able to do the following:

• Understand the different types of AI and how they are applied
• Understand Google’s contribution to the space of AI
• Understand the considerations that are required when preparing to leverage AI
• Describe the Google Cloud solutions for ML and AI

The chapter covers the following topics:

• AI overview and Google’s contribution
• Considerations...

AI overview and Google’s contribution

Google has made meaningful contributions in several areas of the ML and AI space. At a high level, Google has been doing research and development in the ML space for a long time, where they’ve developed frameworks such as TensorFlow, which makes it easier to train ML models. Google also has an extensive library of models that they developed themselves, which can be applied across a variety of data and use cases. One of the most impactful projects in the natural language processing space was the development of the Transformer architecture, the foundation upon which the generative AI revolution was built. What’s interesting about the Transformer architecture is that it is based on the self-attention mechanism, where the model can learn the relationship between words in a sequence without relying on the order of the words. We’ll dive deeper into generative AI models and applications later in this chapter, but it’...

Considerations when building AI models

One of the most important adages of the ML space is garbage in, garbage out. This refers to the fact that if you feed a model data that is not representative of production data, the model will likely not be very accurate or useful. This is important to keep in mind as the instinct may be to feed a model as much data as possible to train the best model. However, when you have trained a massive model, it’s not very cost-effective to serve in production. As you can imagine, the more data you feed a model, the more it’ll cost to train and the larger the model will be when it needs to be served in production. Therefore, in the world of ML, although the first iteration of a model may be based on a larger dataset, future iterations will try to strip away data that doesn’t improve the quality of the model to condense the model into its most cost-effective version. This technique is known as feature engineering. Ideally, this model...

Google Cloud solutions for ML and AI

Google Cloud provides a broad range of solutions to help folks leverage AI regardless of where they are in their data maturity journey. If you are a business user who wants to have a model autocomplete your words and sentences, Gmail or Google Docs may be very valuable. If, however, you are building applications and want to embed AI capabilities within your application, that’s where the Vertex AI platform truly shines.

There are essentially three ways to use ML models on Google Cloud: leverage off-the-shelf models, customize off-the-shelf models with your data and use case, or build models from scratch.

Pre-built models include models for natural language processing, image processing, and structured data processing such as clustering and linear regression. Google also provides the ability to tailor these pretrained models to your use case by training them on your data and classifiers. This is great when one of the pretrained models...

Summary

Google has made considerable contributions to the field of ML and many of the capabilities being launched through Google Cloud are an extension of this work. Where the Google Research or DeepMind teams might’ve historically shared their research exclusively through research papers and eventually open source projects, they can now deliver capabilities directly to Google Cloud customers through the Vertex AI platform and other associated services.

Whether you have a structured data problem to be solved with ML or just need to scrape the information that resides in scanned document files, you will likely find a service or set of services that can help on Google Cloud. Not only does Google Cloud have an extremely robust offering for its infrastructure, foundation models, and APIs, but it also has an extremely mature security posture, which allows it to deliver best-in-class security for any work being done on Google Cloud. In the next section of this book, we’ll...