Overview of Vertex AI

Vertex AI is a machine learning platform that brings together data engineering, data science, and machine learning engineering workflows, enabling teams to collaborate using a common set of tools. The platform offers several options for model training, including AutoML for training models without writing code or preparing data splits and custom training for users who require more control over the training process.

Vertex AI also provides end-to-end MLOps (Machine Learning Operations) tools, which help automate and scale projects throughout the machine learning lifecycle. These tools run on fully managed infrastructure, offering customization options based on performance and budget needs.

The platform can be accessed using a variety of interfaces, such as the Vertex AI SDK for Python, Google Cloud Console, the Google Cloud command line tool, client libraries, and Terraform (with limited support).

Step-by-Step Vertex AI, Understanding ML Workflow

Before we take a deeper look to this, let's understand the machine learning workflow.

After defining the prediction task, the first thing you do is ingest the data, analyze it, and then transform it; then, you create and train the model. Evaluate the model for efficiency and optimization and deploy it to make predictions. For example:

Ingestion, analysis, and transforming are all about data preparation, and you do that through managed data sets within Vertex AI. You have tools to create the data set by importing the data using the console or the API. Whereas, for model training, you have two options: Auto ML or custom; with varying machine learning expertise, for some use cases, Auto ML works wonderfully, such as images or videos and text files. But, if you want more control on your model's architecture, you must use custom models, as they are great for TensorFlow or Pytorch code. Once the model is trained, you have the ability to assess that model and optimize it, and understand the signals behind your models' predictions with explainable AI.

Explainable AI allows you to dive deeper into the model and understand which factors are playing a role in defining what the model is predicting. If you're happy with the model, you can deploy it to an endpoint to serve it for online predictions using the API or the console.

The deployment includes all the physical resources and the scalable hardware that's needed to scale the model for lower latency and online predictions. When the model is deployed, you can get the projections using the command line interface, the console UI, or the SDK and the APIs.

The Machine Learning Workflow with Vertex AI:

• Data Preparation: The first step in any machine learning project is data preparation, which involves extracting, cleaning, and analyzing the dataset. With Vertex AI, you can explore and visualize data using Vertex AI Workbench notebooks, which integrate with Cloud Storage and BigQuery for faster data processing. For handling large datasets, you can utilize Dataproc Serverless Spark from its Workbench notebook.

• Model Training: After preparing the data, you need to train a model using a suitable training method. Vertex AI offers AutoML for training models with tabular, image, text, and video data without coding. For more control over the training process, you can use custom training with your preferred ML framework and hyperparameter tuning options. Vertex AI Vizier can also help optimize hyperparameters for custom-trained models.

• Model Evaluation and Iteration: Once the model is trained, it is essential to evaluate its performance using metrics like precision and recall. You can create evaluations through the Vertex AI Model Registry or include them in your Vertex AI Pipelines workflow. Based on the evaluation results, you can make adjustments to the data and iterate on the model.

• Model Serving: With Vertex AI, you can easily deploy your models into production for real-time online predictions or asynchronous batch predictions. The platform also supports an optimized TensorFlow runtime for serving TensorFlow models at a lower cost and lower latency. In addition, the Vertex AI Feature Store is available for serving features from a central repository for online serving cases with tabular models.

• Model Monitoring: Monitoring the performance of your deployed models is crucial for ensuring their effectiveness. Vertex AI Model Monitoring can help you keep track of training-serving skew and prediction drift, alerting you when the prediction data deviates significantly from the training baseline.

Key Features of Vertex AI

It offers a range of features and tools to support various aspects of the machine learning workflow. Some notable features include:

• AutoML: Develop high-quality custom machine learning models without writing training routines.

• Workbench: A Jupyter-based environment for data scientists to carry out their ML work, from experimentation to deployment and model management.

• Data Labeling: Obtain accurate labels from human labelers for improved machine-learning models.

• Explainable AI: Understanding and building trust in your model predictions with robust explanations.

• Feature Store: A central repository for serving, sharing, and reusing ML features.

• ML Metadata: Track artifacts, lineage, and execution for ML workflows with an easy-to-use Python SDK.

• Model Monitoring: Automated alerts for data drift, concept drift, or other model performance incidents requiring supervision.

• Pipelines: Streamline your MLOps by building pipelines using TensorFlow Extended and Kubeflow Pipelines, with detailed metadata tracking, continuous modeling, and triggered model retraining.

Advantages and Disadvantages

PROS

CONS

Unified Platform

Learning Curve

AutoML and Custom Training

Cost

MLOps tools

Vendor lock-in

Fully-managed infrastructure

Limited support for some features

Multiple interfaces

Potentially overwhelming for small projects

Scalable and flexible

​

Pricing

Vertex AI offers flexible pricing based on model training, predictions, and Google Cloud product resource usage. You can find full pricing rates on the platform's website or estimate your costs using their pricing calculator.

Learning Curve

AutoML and Custom Training

Vertex AI

Build, deploy, and scale machine learning (ML) models faster, with fully managed ML tools for any use case. 

New customers get $300 in free credits to spend on Vertex AI.

Try Vertex AI freeContact sales

• Accelerate ML with a unified data and AI platform and tooling for pre-trained and custom models

• Build generative AI apps quickly and responsibly with Model Garden and Generative AI Studio

• Implement MLOps practices to efficiently scale, manage, monitor, and govern your ML workloads

• Reduce training time and costs with optimized infrastructure

• Learn more in the Vertex AI documentation

      BENEFITS

Build with generative AI

Easily access a variety of foundation models, via developer friendly APIs on Model Garden. Customize, uptrain, and fine tune models to fit your needs with Generative AI Studio. 

Accelerate models to production

Data scientists can move faster with purpose-built tools for training, tuning, and deploying ML models. Reduce training time and cost with optimized AI infrastructure. 

Manage your models with confidence

Remove the complexity of model maintenance with MLOps tooling such as Vertex AI Pipelines, to streamline running ML pipelines, and Vertex AI Feature Store to serve, and use AI technologies as ML features.

KEY FEATURES

One AI platform, every ML tool you need

Choose a model that fits your needs

Jumpstart your ML project with Model Garden, a single place to access a wide variety of APIs, foundation models, and open source models. Kick off a variety of workflows including using models directly, tuning models in Generative AI Studio, or deploying models to a data science notebook. 

End-to-end MLOps

Vertex AI provides purpose built tools for data scientists and ML engineers to efficiently and responsibly automate, standardize, and manage ML projects throughout the entire development life cycle. Using Vertex AI you can easily train, test, monitor, deploy, and govern ML models at scale, reducing the work needed to maintain model performance in production and enabling data scientists and ML engineers to focus on innovation code. 

Data and AI integration

Through Vertex AI Workbench, Vertex AI is natively integrated with BigQuery, Dataproc, and Spark. You can use BigQuery ML to create and execute machine learning models in BigQuery using standard SQL queries on existing business intelligence tools and spreadsheets, or you can export datasets from BigQuery directly into Vertex AI Workbench and run your models from there. Use Vertex Data Labeling to generate highly accurate labels for your data collection.

Low-code and no-code tooling

Vertex AI provides low-code tooling and up-training capabilities so practitioners with a wide variety of expertise can leverage machine learning workloads. With Generative AI Studio, developers can tune and deploy foundation models for their use cases via a simple UI. And, with our off the shelf APIs, developers can easily call upon pre-trained models to quickly solve real-world problems. 

Open and flexible AI infrastructure

Vertex makes it easy to deploy ML models to make predictions (also known as inference) at the best price-performance for any use case. It provides a broad selection of ML infrastructure and model deployment options to help meet all your ML inference needs. It is a fully managed service and integrates with MLOps tools, so you can scale your model deployment, reduce inference costs, manage models more effectively in production, and reduce operational burden.

https://cloud.google.com/vertex-ai#section-1

On May 18, 2021 at Google I/O 2021, Google Cloud announced the first general availability of Vertex AI, a unified AI/ML platform that allows companies and developers to accelerate the development and maintenance of AI/ML solutions. So what exactly is it and why we need to be (or not to be) excited about it.

Advertisements

REPORT THIS AD

Vertex AI

We had two guiding lights while building Vertex AI: get data scientists and engineers out of the orchestration weeds, and create an industry-wide shift that would make everyone get serious about moving AI out of pilot purgatory and into full-scale production

Andrew Moore, vice president and general manager of Cloud AI and Industry Solutions at Google Cloud

Vertex AI is a managed machine learning (ML) platform that allows companies to accelerate the deployment and maintenance of artificial intelligence (AI) models.

Advertisements

REPORT THIS AD

According to Google, Vertex AI requires nearly 80% fewer lines of code to train a model versus competitive platforms, enabling data scientists and ML engineers across all levels of expertise the ability to implement MLOps to efficiently build and manage ML projects throughout the entire development lifecycle.

Vertex AI – Unified AI/ML UI, now in general availability on Google Cloud Platform

Why there is a need?

Common building construct of any machine learning frameworks today consists of:

• Data Extraction

Extraction of data using standard ETL or ELT tools or services and storing them in an object storage or any database for further analysis.

• Data Analysis, Preparation and Cleaning

Once the data is stored on cloud, a standard process of analysing the data and also transforming the data comes into play. This process can either be carried out by using standard ETL/ELT tools or using cloud services.

• Model Training

Once the data is cleaned and transformed, Data scientists train the model using a bunch of ML algorithms. The algorithms can be either Custom models or AutoML depending on the dataset processed and stored in the previous steps.

• Model Evaluation and Tuning

Upon training, models are evaluation for prediction accuracy and bias. This is a continuous process of development and requires to train the model again to achieve a desired result.

• Model Deployment

Once the model are trained and evaluated, it is deployed on a scalable infrastructure with endpoints exposed.

• Model Prediction and Monitoring

A critical step is to monitor the performance of the models on live dataset. Based on the accuracy rate of the prediction, the models are retrained and the entire process starting from model training till monitoring are carried out.

Advertisements

REPORT THIS AD

Having said that, the above pipeline stages are quite significantly different to each others. Depending on how you do your ETL, rest of the pipeline becomes quite different. For e.g. tabular CSV dataset ML pipeline execution is quite different than an image classification one. The storage location and type of your dataset is also a big factor e.g. store in local SSD disk or on Cloud Storage. Hence there is a need to have a unified definition of data.

Now let’s look at the Model building and training stages, using TensorFlow as an ML framework is different from using PyTorch, AutoML, etc. The same goes for the deployment of such frameworks into production. Hence there is also a need to ensure every model framework are treated in the same way.

Advertisements

REPORT THIS AD

Vertex AI embraces MLOps technique and re-architect the entire Model pipeline into one single roof. This unified framework brings in a lot of excitement among the AI/ML developers and is also quite unique.

It lowers the barrier for MLOps, the creation and deployment of AI applications, and, ultimately, AI-fueled digital transformation. Before this, there were no good solutions out there that provided all the professional-data scientist-and-data engineer-grade capabilities in an integrated platform. Enterprises had to stitch together a stack of disjointed tools and technologies making it hard for nearly all enterprises to get the full value from their AI initiatives.

Forrester Analyst Kjell Carlsson on Vertex AI

Key Features

Google’s Vertex AI comes into the picture providing an unified definitions and implementation of the below concepts:

• Dataset

A dataset can be structured or unstructured. It has managed metadata including annotations, and can be stored anywhere on GCP. At present only Cloud Storage and BigQuery are supported.

• Feature Store

Vertex Feature Store, a repository to help data science practitioners organize, store, and serve machine learning features

• Labeling

Vertex data labeling service provides an end user with an easy out of the box solution to label their data.

• Containerised Training Pipeline

A training pipeline is a series of containerised steps that can be used to train an ML model using a dataset. The containerisation helps with generalisation, reproducibility, and auditability.

• Continuous Monitoring

Vertex Model Monitoring, a self-service model tool that lets users monitor the quality of machine learning models over time.

End to End Unified AI Platform – Vertex AI

COMPLETE LIST OF VERTEX AI FEATURES ARE IN HERE.

Hence the big idea is regardless of data or model framework, the pipeline is unique across as the same. So, once you create a dataset, you can use it for different models. You can get Explainable AI from an endpoint regardless of how you trained your model.

Advertisements

REPORT THIS AD

TRY OUT VERTEX AI IN GOOGLE CLOUD PLATFORM.