Use Artificial Intelligence (AI) or the AI Platform to train machine learning at scale. It can also be used to host and train existing models on Google Cloud, and to make predictions about new data using your own model types. These ideas and procedures will undoubtedly be beneficial to your company’s big data processing.
What role does the AI Platform play in the Machine Learning process?
The diagram below illustrates the stages of the Machine Learning workflow. The AI Platform delivers managed services and APIs in the blue boxes:
The AI Platform can be used to manage the following steps in a Machine Learning workflow, as shown in the diagram:
On your data, train Machine Learning models:
- Training model (Train Model)
- Evaluation model accuracy (Evaulate Accuracy Model)
- Tune hyperparameter
- Monitor predictions continuously.
- Manage your models and versions of models.
- Apply your trained model.
- Prediction model request to online
- model Batch prediction model request (only for TensorFlow)
Components for AI training
The following is a list of the components that make up the AI Platform, along with their primary functions:
AI training services Platform AI Platform
Models can be trained utilizing a variety of options and adjustments thanks to training services. To support your training effort, you may choose from a variety of engine types, enable distributed training, employ hyperparameter tuning, and accelerate with GPU and TPU.
You can also modify your training application in a variety of ways.
algorithmbeta. If the built-in algorithms aren’t suitable for your needs, you can either ship your own training app or develop a bespoke container containing your training app and its dependencies to operate on the AI Platform.
services for forecasting Prediction services on the AI Platform allow you to deliver predictions based on learned models, whether or not the models were trained on the AI Platform.
service You can utilize the AI Platform Data Labeling Service (beta) to request labels for the data sets you’ll be using to train custom machine learning models. You can ask for your video, image, or text data to be labeled.
To submit a labeling request, you must include a sample of labeled data, a list of all possible labels for your data set, and instructions on how to apply the label. When the human labeler completes the labeling request, you will receive an annotated data set that can be utilized to train machine learning models.
Here are the most popular ways to interact with the AI Platform. This section explains how to interface with the AI Platform using the following tools:
- The Google Cloud Console allows you to upload models to the cloud and manage your models, versions, and tasks. This option provides a user interface via which you may interact with your machine learning resources. Your AI Platform resources are connected to Google Cloud capabilities like Cloud Logging and Cloud Monitoring as part of the Google Cloud.
- With the gcloud ai-platform command line functionality, you can manage models and versions, submit tasks, and execute other AI Platform operations via the command line. For most AI Platform operations, Google suggests the gcloud command, and for online prediction, the REST API.
- The AI Platform REST API provides RESTful services for managing tasks, models, and versions, as well as making predictions with Google Cloud models.
To access the API, you can use the Google APIs ClientLibrary for Python. When you utilize client libraries, you’re dealing with the Python representation of the API’s resources and objects. Working with HTTP requests directly is more difficult and requires more code.
For offering online predictions, Google suggests using the REST API.
User-managed notebooks in Vertex AI Workbench User-managed notebook instances in Vertex AI Workbench allow you to construct and maintain pre-packaged JupyterLab virtual machines (VMs).
A set of deep learning packages, including support for the TensorFlow and PyTorch frameworks, is pre-installed on user-managed notebook instances. You can set up instances that are CPU-only or GPU-enabled.
Google Cloud authentication and permission safeguard user-managed notebook instances, which can be accessed via the user-managed notebook instance URL. Notebook instances owned by users are also linked to GitHub and can be synchronised with GitHub repositories.
Deep Learning VM for AI Platform
Deep Learning VM Images are virtual machine image collections that are tailored for data science and machine learning operations. Machine learning frameworks and technologies are preloaded. To speed up data processing activities, you can use it directly on instances with GPUs.
Many different frameworks and processors are supported by Deep Learning VM images. TensorFlow Enterprise, TensorFlow, PyTorch, and high-performance generic computing are presently supported by pictures, with variants for CPU and GPU-only workflows.