Vertex AI Guide: How to Build ML Models

As of January 23rd 2023, Google’s AI Hub is officially deprecated. What this means is that you’ll no longer be able to use their services after January 23rd, 2024. 

But don’t worry! You’ll find that they’re migrating over to Vertex AI, which includes all functionality of AI Hub, as well as Google’s AutoML (auto-machine learning) platform. 

The landscape of AI technology is constantly advancing at a rapid rate, so it’s more than likely that we’ll see these services change and update with time. Eventually, they too will be replaced. But for now, it’s useful to make the most of them. 

So let’s get into what Vertex AI is, how it differs from Google AI Hub, and how it can help you and your business. 

The logo looks like a book opened on its back with its pages at a 95 degree angle. Blue dots and dashes like morse code are floating above it

Key concepts you need to know

Before we get totally engrossed, it’s important to know what many of the terms and phrases used throughout this article mean. The field of artificial intelligence is enormous, so it’s useful to have a reference point to come back to if you’re unsure of how a certain piece of tech fits in with the rest of the puzzle.  

Docker

A software platform used by developers for shipping applications, that lets you build, test, and deploy these applications quickly and easily. The software is bundled into groups known as containers. 

Docker images are read-only templates that show instructions for how a container can be put together. 

Containers

A standard unit of software that provides an isolated context where an app and its environment can run. It contains all of the necessary components to operate anywhere. This includes libraries, data science, system tools, source code, and runtime.

Infographic of how docker containers work

Source

Pipeline

The way in which the operation from your business is streamlined and organized. Datasets from your company’s workflow and lifecycle, and customer information are processed through containers, each of which have a list of tasks and parameters to help sort and transform the information, train new ML models, and optimize how your business runs. 

Kubernetes 

Similar to Docker, Kubernetes is a platform that can run thousands of containers in production. Groups of containers are known as pods, which you have the ability to replicate. It uses labels to identify various containers and pods, and can query based on these labels. It connects these pods to the network and wider online ecosystem. 

Kubeflow Pipeline 

Kubeflow is an end-to-end open source ML toolkit operated by Kubernetes. You can use it to build and deploy moveable, transformable (scalable) ML workflows, and to manipulate components in a pipeline, or the entire pipeline itself. 

TensorFlow

TensorFlow is an end-to-end platform with a deep learning framework for machine-learning that lets you: 

  • Build your own ML models with the TensorFlow ecosystem
  • Deploy modules over various different platforms, such as on servers, browsers (include Microsoft Edge), mobile, CPUs and GPUs
  • Use automatic differentiation, training loops, and variables which work through data for successful ML outcomes

What was the Google Cloud AI Hub?

Google AI Hub was a part of Google Cloud, which let you build and host applications, and store data on their computer network infrastructure.

It was a collection of AI tools, plugins, and assets which you could use to create an effective, automatic pipeline for your business. You could teach your own ML algorithms, or collaborate with others in your organization to code and share your AI artifacts (all digital products that are used in an AI Tool), and keep them all safe with Google Cloud storage. 

These tools could be powered directly by Google, or through other popular frameworks such as Kubeflow or Jupyter Notebooks (a web-based interactive computing platform, part of Project Jupyter).

Old google ai hub desktop homepage

Source

Here are a few examples of what was also on offer as part of the Cloud system, which have now been integrated into Vertex AI:

  • AutoML: Ideal for both AI beginners, or seasoned experts. It allows you to create your own custom machine learning models that are tailored to your business needs which you can then integrate into your business.
  • Vision AI: An API (application programming interface) that analyzes images and extracts key information from them, such as object detection, iconic landmarks, text-search. You could also use your own custom models or AutoML to train image classification algorithms. 
  • Cloud Natural Language: Analyzes the entities, syntax, and sentiment of a piece of text, and categorizes it into its relevant field, such as technology. The Translation API also allows you to conduct this process across multiple languages to help your global consumer base.

What is Vertex AI?

Vertex AI was initially released back in 2021, and was designed to unify AutoML and custom ML applications, creating a more accessible environment for people to experiment and modify their AI business practices. Now, the entirety of Google’s AI Hub is migrating over to Vertex AI. 

A red box that reads "AI Hub is deprecated and will no longer be available on Google Cloud after January 23, 2024. All the functionality of AI Hub and new features are available on the Vertex AI platform."

Its biggest competitors are companies such as Amazon Web Services (AWS) and their SageMaker, which is already a very comprehensive AI model. Google Cloud Platform (GCP) combining key elements of its AI tech into one cohesive structure within Vertex AI may give it an edge in the market. 

But what, exactly, does Vertex AI do?

To understand that in more detail, Google’s Developer Advocate, Priyanka Vergadia, is here to help. 

If you’re unable to watch the video, here’s a brief breakdown. 

In short, Vertex AI works to simplify the process of developing and training machine learning, and gives you more options to modify your pipeline exactly how you want to. Generally, the process of training an AI operates using the following steps:

  1. Ingest
  2. Analyze
  3. Transform
  4. Train
  5. Model
  6. Evaluate
  7. Deploy
  8. Predict

Vertex AI provides open-source tools that lets you control every element. 

  • You can manage and label the datasets.
  • Train either AutoML or custom models which work well with different AI platforms. With AutoML, Vertex AI will take care of writing any code you need. The custom models are highly compatible with platforms such as TensorFlow. 
  • You can then go about optimizing these models and understanding them with Explainable AI.
  • With your model trained, you are ready to deploy it. Vertex AI has scalable hardware which makes your systems accessible for lower latency and online predictions.
  • Then, you can collect batch predictions your AI makes using the command line interface, Google Cloud console user interface, software development kit, or the APIs.  

Migrating to Vertex AI

The migration process is free, and it creates a copy of your AutoML and AI Platform datasets and models, provided that they are younger than 18 months old. Since AutoML models have a maximum lifespan of 18 months, Vertex AI cannot migrate anything past that point. 

It is worth noting that the individual AutoML and AI Platform elements will still be available without unification, but “future improvements will be implemented on Vertex AI.”

Vertex AI’s main features

Before you migrate, you might want to know exactly what Vertex AI offers. So let’s take a quick look at its main features.

Access to Gemini

Gemini is a multimodal model that has incredible capabilities. Its model outputs include just about anything, and you can input almost any kind of training data. You can prompt and test your custom model with any kind of content, including:

  • Images
  • Videos
  • Text
  • Code

Access to this model is a developer’s dream, so it’s well worth migrating.

Access to hundreds of models

Model Garden is like a model registry, full of over 130 different model types. Here, you can discover the best one for you and your business, try them out, and even start a custom training process so they are optimum for your needs. Google offers multiple tuning options making the model training job as easy as possible.

Alongside these models are fully formed tools. This means you can prototype and deploy them into your applications.

Open and integrated platform

The Vertex AI dataset includes training and prediction features, which means you can take full advantage of open-source frameworks and infrastructures to get building quickly.

There are multiple integrations throughout the platform designed to make every step of the process simpler, from training data preparation to model training.

Search and conversation

If you’re looking to build a search or chat application, like a chatbot, then you’re in luck. Vertex AI Search and Conversation means developers can quickly train models, build their application and deploy it to their target audience.

Conclusion 

It’s difficult to say at this moment in time whether Vertex AI will remain Google’s go-to AI software, or if they have something new in the works. AI Hub didn’t seem to be around for that long before it was integrated, but that’s the nature of AI technology. 

For now, there are a plethora of tutorials and help-guides on how to use Vertex AI and make the most out of your AI experience, whether you are a long-time data scientist, a business owner, or a complete beginner looking to start your first Google Cloud project. 

For more news and all the latest updates about the world of AI, stay in touch with Top Apps.