Home BigQuery Efficient ML model management and deployment with Google Vertex AI: From the model registry to vector search

Efficient ML model management and deployment with Google Vertex AI: From the model registry to vector search

ML model management and deployment with Google Vertex AI Visual

Google Vertex AI enables the efficient development, deployment and management of machine learning models. The wiki provides a guide for a comprehensive overview of the platform's key features, including model registry, online and batch predictions, and vector search.

The Google Vertex AI tool is crucial for optimizing ML workflows, accelerating data processing and improving the accuracy of results. The platform is particularly suitable for scaling and managing AI projects in various business areas.

Table of contents

1. enable APIs and pre-built models: Optimal use of Google Vertex AI

 

1.1 Dashboard 

Vertex AI is a comprehensive platform that provides a variety of APIs to enable the use of artificial intelligence in different use cases. To fully utilize Vertex AI, the following APIs must be activated:

1.
Vertex AI API:

Core API for accessing the main functions of the platform.

2.
Cloud Storage API:

Enables the storage and retrieval of data required for machine learning models.

3.
Notebook API:

Supports the creation and management of Jupyter notebooks.

4.
Dataflow API:

Enables the processing of large amounts of data in real time or in batches.

5.
Artifact Registry API:

Manages and saves build artifacts such as containers.

6.
Data Lineage API:

Enables the tracking of data origin to improve data quality and governance.

7.
Data Catalog API:

Helps to organize and manage the metadata of data resources.

8.
Compute Engine API:

Enables the management and scaling of compute resources.

9.
Dataform API:

Supports the management of data pipelines and ETL processes.

10.
Vision AI API:

Specialized in image recognition and related visual machine learning tasks.

In addition, Vertex AI enables the monitoring of specific metrics of individual services to effectively monitor performance and usage. This provides valuable insights into the operational aspects of the platform and helps to optimize the applications.

 

In-depth insights into Google Vertex AI

The white paper provides comprehensive information on how to efficiently manage and deploy ML models with Google Vertex AI. It covers key topics such as the Model Registry and Vector Search and shows how these features help to optimize AI projects. An in-depth reading enables a better understanding of the possibilities and benefits of Google Vertex AI in practice.
Screenshot of the Vertex white paper

1.2 Model Garden 

Model Garden is a collection of pre-built machine learning models and tools available in Google Cloud Vertex AI. It enables developers and data scientists to quickly and easily create, test and deploy models for a variety of use cases. A total of 153 models are available.

The models are divided into three main categories:

  • Broad knowledge: Versatile models that have been trained on large amounts of data.
  • Adaptable: Can be optimized and specialized for different tasks.
  • Specialization: Refined versions of Foundation Models, adapted to specific tasks.
  • Advanced customization: Require additional data and code for precise customization and performance enhancement.
  • Ready to use:Prefabricated models for specific tasks, ready to use.
  • Possible fine-tuning: Some solutions allow adjustments with your own data for individual optimization.

The models can also be divided into further categories:

  • Language Models
  • Tabular Models
  • Document Models
  • Multimodal Models
  • Vision Models
  • Speech Models
  • Video Model

 

Another way to categorize models is according to their features:

  • Vertex AI Studio (ready-to-use models that can be tested in the console, such as Gemini)
  • API available (models that can be called via an API)
  • Open source
  • Notebook support
  • Pipeline support
  • One-click deployment

2. overview of the most important models and APIs in Google Vertex AI: possible applications and use cases

Google Vertex AI includes a wide range of models and APIs that cover various AI use cases. This overview provides insights into the use of tools such as Gemini 1.0 Pro and AutoML Vision to optimize ML workflows and efficiently scale AI projects.

2.1 Language:

Gemini 1.0 Pro: Gemini 1.0 Pro is the name of the large language model Gemini, which understands language and is a base model suitable for a variety of natural language processing tasks, such as summarization, instruction following, content generation, sentiment analysis, entity extraction, classification, etc. The content that Gemini 1.0 Pro can generate includes summaries of documents, answers to questions, labels for classifying content and much more.

Use Cases: 

  • Question answer
  • Classification
  • Sentiment analysis
  • Entity extraction

Average Word Embedding Classifier: The Average Word Embedding Classifier is a simple, lightweight text classification model that uses average word vectors to create a text representation and evaluates it via a two-layer network with RELU activation. The model is designed for use on devices and can be used with MediaPipe Tasks TextClassifier on various platforms such as Android, iOS, web and desktop.

Use Cases:

  • Text classification

Text moderation: Text moderation analyzes a document and identifies harmful and sensitive categories that apply to the text found in the document.

Use Cases:

  • Checking for potentially offensive content that could lead to a negative perception of the brand.
  • Ensure that user-generated and editorial content is brand safe for advertising.
  • Identification of toxic comments in forums and chat messages that could offend and deter users.

2.2 Tabular:

Claude 3 Opus: Claude 3 Opus from Anthropic is the company's most powerful AI model and demonstrates outstanding capabilities in complex tasks. The model can handle open-ended tasks and unfamiliar scenarios with impressive fluency and human-like understanding. All Claude 3 models can process images, output text and have a 200K K

Use Cases:

  • Task automation: planning and execution of complex actions via APIs and databases, interactive P
  • Research and development (R&D): Review of research work, idea generation and hypothesis generation, drug research.
  • Strategy: Advanced analysis of charts and graphs, financial and market trends, forecasts.
  • Visual skills: Image and word processing to analyze and understand charts, graphs, technical drawings, reports and other visual media.

AutoGluon: AutoGluon automates machine learning processes to achieve high predictive performance in your applications. However, this version only supports tabular data, which allows you to train and implement highly accurate machine learning and deep learning models on such data. With a simple call to fit(), you can achieve high accuracy in standard supervised learning tasks (classification and regression) without the complexity of data cleaning, feature extraction, hyperparameter optimization and model selection.

Use Cases:

  • Prediction of tabular data: Predicting the values of a target column based on the other columns of a table data set.

AutoML E2E: The "Tabular Workflow for End-to-End AutoML" is a complete AutoML pipeline for classification and regression tasks, similar to the AutoML API, but offers more control over the individual process steps. The pipeline includes automated feature engineering, architecture search and hyperparameter tuning. Successful models go through several stages, including cross-validation and model combination, to optimize the final model selection. The AutoML solution was successfully tested in a competition and won second place.

Use Cases:

  • Marketing and customer analysis: estimation of order frequency, probability of customer churn, conversion probability of leads, customer lifetime value, campaign allocation.
  • Resource utilization: probability of equipment failure, estimation of supply and demand for drivers, staff turnover, life expectancy of equipment.
  • Risk management: estimation of the amount and probability of loss, probability of fraud, probability of default.
  • Ranking: Optimal product placement and advertising effectiveness.

2.3 Document:

Claude 3 Opus: see following section

Document AI OCR Processor: Document OCR recognizes and extracts text from documents in over 200 printed and 50 handwritten languages. It identifies text blocks, paragraphs, lines, words and optionally symbols in PDFs and images and can automatically straighten documents to increase accuracy. In addition, it enables the recognition of font styles, language cues and the evaluation of image quality for optimized further processing.

Use Cases:

  • Document digitization: Digitize text from documents, automate data entry, improve and verify data quality, programmatic pre-processing.
  • Invoice and claims processing: Extraction of data from invoices to automate invoice and claims processes and reduce errors.
  • Document search & Q&A: Extract data from documents to enable automated processes for questions and search functions.
  • Contract management: Extract data from contracts to increase their accessibility and searchability and to support the correct execution of contracts.
  • Archiving: Converting paper documents into electronic formats to improve document accessibility.
  • Compliance: Extract data from documents to ensure compliance and perform content-based moderation.

2.4 Multimodal:

Gemini 1.5 Pro: The Gemini 1.5 Pro is a versatile base model that is particularly suitable for multimodal tasks such as visual understanding, classification, summarizing and generating content from images, audio and video. It efficiently processes visual and textual input such as photos, documents, infographics and screenshots. The model is part of the Gemini model family, which offers different sizes and capabilities, including specialized versions for text and visual content.

Use Cases:

  • Visual information search: Combination of external knowledge with information extracted from images or videos to answer questions.
  • Object recognition: Detailed identification of objects in images and videos.
  • Digital content understanding: Extraction of information from visual content such as infographics and websites.
  • Structured content production: Generation of responses based on multimodal input in formats such as HTML and JSON.
  • Description and captioning: Creation of descriptions for images and videos in various languages.
  • Conclusion: Novel information creation without memorization or retrieval.
  • Audio: Analysis of voice files for summaries, transcriptions and Q&A.
  • Multimodal processing: Simultaneous processing of different media types such as video and audio.

2.5 Vision, speech and video models 

These are included in the white paper.

3. notebooks for data science and machine learning

Notebooks are interactive environments that enable the integration of code, text and visualizations in one document. These environments are often used in the fields of data science and machine learning, particularly for exploratory data analysis, prototyping models and communicating results.

3.1 Colab Enterprise

Colab Enterprise is a Google Cloud Platform-based notebook environment designed specifically for enterprises to provide a scalable and secure platform for collaborating on data science projects. This environment supports the use of powerful GPUs and CPUs to meet the requirements of different projects. Colab Enterprise also provides security mechanisms for data processing and storage and enables real-time collaboration of multiple users on a project. In addition, seamless integration with other Google Cloud services, such as BigQuery and Cloud Storage, is possible.

3.2 Workbench

Workbench is an open source notebook platform based on Jupyter that can be operated both locally on a computer and in the cloud. The platform is highly customizable and extensible through a variety of plugins and extensions. It offers a flexible working environment that can be tailored to the specific needs and preferences of data scientists. As the Workbench is an open source project, the source code is freely available and can be used and modified by users free of charge.

4th AI Studio

4.1 Overview

Vertex AI Studio is a web-based development environment for machine learning (ML) on the Google Cloud Platform (GCP). It enables teams of data scientists, ML engineers and business users to create, train and deploy ML models without writing code.

Vertex AI Studio offers a range of features that simplify the development of ML models, including:

  • A drag-and-drop interface forcreating ML pipelines
  • Pre-trained models forvarious tasks such as image classification and natural language processing
  • Tools for training and optimizing models
  • Options for providing models in production

Vertex AI Studio is a good choice for teams that want to develop and deploy ML models quickly and easily. It's also a good choice for teams new to ML, as it offers a user-friendly interface and a range of pre-built features.

 

4.2 Multimodal

Vertex AI Studio supports the development of multimodal ML models that can process data from different modalities such as image, text and audio. This enables the development of more powerful and versatile ML models.

Vertex AI Studio offers a range of functions that support the development of multimodal models:

  • Data pre-processing tools formerging and preparing data from different modalities
  • Models for multimodal tasks such asimage-text matching and audio-video recognition
  • Tools for evaluating the performance of multimodal models

Vertex AI Studio is a good choice for teams that want to develop multimodal ML models. It offers a range of features that facilitate the development and deployment of these models.

 

4.3 Language

Vertex AI Studio supports the development of language models that can process and generate text. This enables the development of applications such as chatbots, machine translation and text generation. Vertex AI Studio offers a range of functions that support the development of language models:

  • Pre-trained language models fordifferent languages
  • Tools for training language models on user-defined data
  • Tools for evaluating the performance of language models

Vertex AI Studio is a good choice for teams that want to develop language models. It offers a range of features that facilitate the development and deployment of these models.

4.4 Vision

Vertex AI Studio supports the development of vision models that can process images and videos. This enables the development of applications such as image classification, object recognition and image segmentation.

Vertex AI Studio offers a range of functions that support the development of image processing models: 

  • Pre-trained vision models forvarious tasks such as image classification and object recognition
  • Tools for training vision models on user-defined data
  • Tools for evaluating the performance of vision models

Vertex AI Studio is a good choice for teams that want to develop vision models. It offers a range of features that facilitate the development and deployment of these models.

 

4.5 Speech

Vertex AI Studio supports the development of speech models that can process and generate speech. This enables the development of applications such as speech recognition, speech synthesis and speech translation. Vertex AI Studio offers a range of features to support the development of speech models, including:

  • Pre-trained speech models fordifferent languages
  • Tools for training speech models on user-defined data
  • Tools for evaluating the performance of speech models

Vertex AI Studio is a good choice for teams looking to develop speech models. It offers a range of features that facilitate the development and deployment of these models. Note: Speech is not yet fully supported in the current version of Vertex AI Studio. However, it is under development and will be available soon.

5. efficient data management for ML models

5.1 Feature Store

The Feature Store is a centralized location for managing features for machine learning (ML) models. It allows data scientists and ML engineers to store, version and reuse features in one place, simplifying the development and deployment of ML models. The Feature Store offers a range of features that simplify feature management, including: Storage of features in multiple formats, including numeric data, categorical data, and text data Versioning of features to track changes to features over time Reuse of features in multiple ML models Search and filtering capabilities to find the features you need Integration with Vertex AI to simplify ML model development and deployment.

The Feature Store is a good choice for teams that use multiple ML models and want to manage their features centrally. It can help speed up the development and deployment of ML models and improve the quality of ML models.

  • Storage of features in various formats, including numerical data, categorical data and text data
  • Versioning of features to track changes to features over time
  • Reuse of features inmultiple ML models
  • Search and filter functions tofind the required features
  • Integration with Vertex AI to simplify the development and deployment of ML models

The Feature Store is a good choice for teams that use multiple ML models and want to manage their features centrally. It can help speed up the development and deployment of ML models and improve the quality of ML models.

In-depth insights into Google Vertex AI

The white paper provides comprehensive information on how to efficiently manage and deploy ML models with Google Vertex AI. It covers key topics such as the Model Registry and Vector Search and shows how these features help to optimize AI projects. An in-depth reading enables a better understanding of the possibilities and benefits of Google Vertex AI in practice.
Screenshot of the Vertex white paper
Start your AI revolution with Google Vertex AI

Discover the full potential of Google Vertex AI for your business. Our experts are ready to help you implement this powerful platform and develop customized AI solutions for your specific challenges.

Arrange a free consultation now and find out how you can use Google Vertex AI:

  • Optimize your ML workflows
  • Accelerating the development of AI models
  • Making data-driven decisions
  • Give your company a competitive edge

Let's shape the future of your company with AI together. Contact us today and take the first step towards AI-supported innovation!

Published by:

Christopher Maier

Google Cloud Platform (Cloud Infrastructure | Cloud Solutions) Consultant

author

How did you like the article?

How helpful was this post?

Click on a star to rate!

Average rating 3.7 / 5.
Number of ratings: 6

No votes so far! Be the first person to rate this post!

INFORMATION

More information

The connection between SAC and DSP

Seamless planning combines SAC and DSP

The seamless integration of SAP Analytics Cloud (SAC) and SAP Datasphere (DSP) optimizes, accelerates and simplifies your planning processes. So...
Wiki project management

3 steps to successful stakeholder management

Stakeholder management is a key success factor for any project. It...
Business Content in the SAP Datasphere - Wiki Visual

Business content in the SAP Datasphere - best practices for rapid progress

SAP Datasphere offers predefined business content, similar to SAP BW or SAP Analytics Cloud. This consists of data models...

Optimized vs. classic mode in the SAP Analytics Cloud

The SAP Analytics Cloud (SAC) currently offers users two main modes...
Wiki_Banner_Using Sharepoint as a Datasource_2

SharePoint as a data source for BW/DS

SharePoint is a widely used platform in companies and offers...
What can the new SAC Add-in for Microsoft PowerPoint (Composable Analytics) do? Visual

What can the new SAC Add-in for Microsoft PowerPoint (Composable Analytics) do?

This wiki article is about the new SAP Analytics Cloud (SAC) add-in for Microsoft PowerPoint, which provides a...