Vertex AI

The AI Platform used to be a one-stop shop on Google Cloud Platform (GCP) for enterprise needs related to machine learning (ML). It comprised model training at scale, model hosting, experimentation, generating predictions and more. Since the launch of Vertex AI, however, this newer Google system has replaced the AI Platform, and according to GCP it is the next generation of the AI Platform based on a series of new features. It is recommended to migrate existing resources from the AI Platform to Vertex AI to get the latest ML platform features, production models using MLOps, and simplify the end-to-end ML lifecycle.

Vertex AI can be used for the following components in an ML workflow (the list is not exhaustive):

  • Store the training data
  • Experimentations
  • Train the ML model
  • Evaluate the ML model
  • Hyperparameters tuning
  • Store the trained model
  • Deploy the trained model
  • Batch and online predictions
  • Monitor the deployed model
  • Lineage tracking, etc

Features of Stacktics’ Vertex AI Implementation

Vertex AI Data Labeling: Vertex AI Data Labeling jobs are very helpful when there is a need to do data labeling. This service lets you request human labeling for labeling image, video, or text data required for model training. A representative labeled data sample is required, which includes all possible labels for a given dataset, along with instructions on how to assign labels.


Vertex AI Datasets: Vertex AI Datasets serve all the data needed for training AutoML or custom models. Users can create an empty dataset and upload/import the data inside the created dataset. Data can be imported from the local machine, BigQuery, Google Cloud Storage, etc. Later, this dataset can be fed for model training.


Vertex AI Feature Store: Vertex AI Feature Store is a fully managed solution providing a central repository to store, organize, and serve ML features. The central feature repository empowers an organization to discover, share, and re-use ML features which can lead to fast development and deployment of ML systems.


Vertex AI Experiments: Vertex AI Experiments enable you to track and compare steps (preprocessing, transformation, training, etc.), inputs (hyperparameters, datasets, algorithms, etc.), and outputs (trained models, metrics, checkpoints, etc.) of ML experiments. This is extremely useful to figure out which experiment worked best and plan the next steps in the experimentation journey.


Vertex AI Tensorboard: Vertex AI Tensorboard is a managed and enterprise-ready version of open-source TensorBoard. It helps to visualize, track, and compare models and experiments along with the ability to share with team members.


Vertex AI Workbench: Vertex AI Workbench is a Jupyter notebook-based development environment for the end-to-end data science workflow. Vertex AI Workbench can be used to access the data, preprocess the data, train the model, generate predictions, and much more without leaving the Jupyter interface.


Vertex AI for Training: Vertex AI supports AutoML and custom model training methods.


Vertex AI Pipelines: Vertex AI Pipelines help to orchestrate, automate, monitor, and govern end-to-end machine learning workflows in a serverless manner.


Vertex AI Vizier: Vertex AI Vizier is a black-box optimization service that helps tune hyperparameters for machine learning models.


Vertex AI Model Registry: Vertex AI Model Registry is a model repository that allows the management of the entire lifecycle of ML models. Model Registry helps access data regarding custom models, AutoML models, and imported (trained outside Vertex AI) models.


Vertex AI for Prediction: Vertex AI supports online prediction and batch prediction methods.


Vertex Explainable AI: Vertex Explainable AI helps to assess and understand the model outputs for regression and classification tasks. It helps to understand the contribution of each feature toward the final predicted result. This information can be used to understand the model behavior and decide on ways to improve the model performance.


Vertex ML Metadata: Vertex ML Metadata helps you store and analyze the metadata and artifacts produced by ML pipeline/system to understand, debug, and audit the performance of the ML pipeline/system or the artifacts produced.


Vertex AI Model Monitoring: Vertex AI Model Monitoring helps to maintain the performance of the ML model in production by monitoring the incoming data for prediction to identify skew and drift. Vertex AI Model Monitoring provides support for feature skew and drift detection for numerical and categorical features.


Vertex AI Matching Engine: According to GCP, “Vertex AI Matching Engine provides the industry’s leading high-scale, low-latency, vector similarity-matching (also known as an approximate nearest neighbor) service, and industry-leading algorithms to train semantic embeddings for similarity-matching use cases.”

Using the components mentioned above, Stacktics can help you create robust and production-grade ML systems/pipelines based on your custom use case and needs. Submit the Contact Us form and we will connect to help you unleash the power of the Vertex AI Platform for your machine learning requirements.

Have a question, get an answer. We would be happy to chat.