Introduction to AI Hub
AI Hub offers a collection of components for developers and data scientists building artificial intelligence (AI) systems. You can use the AI Hub to:
- Find, deploy, and use Kubeflow pipelines and components.
- Explore code and learn in interactive Jupyter notebooks.
- Explore and reuse TensorFlow modules.
- Explore, deploy, and use trained models.
- Use prepackaged virtual machine (VM) images to quickly set up your AI environment.
- Share AI components with your colleagues.
An asset is an AI resource that has been published to the AI Hub. AI Hub organizes assets by category to make assets easier to find and use.
Below are some of the asset categories available on AI Hub.
Kubeflow pipelines and components
Kubeflow pipelines are end-to-end machine learning (ML) workflows based on containers. Pipelines are built from components which are self-contained sets of code, packaged as container images. Each pipeline component performs a step in the ML workflow, such as preprocessing, data transformation, or training a model.
The Kubeflow Pipelines system orchestrates the execution of pipelines, creating and running component containers in the order defined by the workflow graph. You can use the Kubeflow Pipelines system to:
- Build and share repeatable ML workflows.
- Create ML experiments and move them to production.
For more information, see the guide to understanding Kubeflow pipelines and components.
Jupyter notebooks provide an interactive coding and learning environment. You can view explanatory content, modify and run sample code, and view the results. Learn more about discovering notebooks on AI Hub, or uploading a notebook so that you can share it with your colleagues.
A TensorFlow module is a reusable piece of a TensorFlow graph. With transfer learning, you can use TensorFlow modules to preprocess input feature vectors, or you can incorporate a TensorFlow module into your model as a trainable layer. This can help you train your model faster, using a smaller dataset, while improving generalization.
For example, follow the image retraining notebook on Colaboratory to learn how to use a TensorFlow module to retrain an image classification model.
Trained models are a representation of what an ML system has learned from its training data. Models on AI Hub can be reused to preprocess input feature vectors, or deployed as ML services. You can download a model from AI Hub to experiment with it locally, or you can deploy a model to AI Platform for predictions. To learn more about deploying a trained model as an ML service, read the AI Platform Prediction overview.
Services are APIs and other building blocks that you can use to add ML functionality to your code. Here are some ways that these services can help you:
- Cloud Translate, Vision, and Natural Language APIs add features to your code by interacting with pretrained models on proven datasets.
- You can use AutoML services to train your model using your own dataset with the benefits of transfer learning. Transfer learning can help you train a model faster, use a smaller dataset, and improve generalization.
The technical guides on AI Hub demonstrate how to implement complex AI use cases. These guides combine services, infrastructure, and custom code to implement complete solutions. By following the technical guides on AI Hub, you learn best practices that you can apply when designing an AI system. For more information, see how to find and learn from technical guides.
The VM images on AI Hub come preconfigured with the frameworks and tools that are necessary for ML workloads. By starting with a VM image, you can quickly create an environment that has been optimized for ML with the option to use Cloud TPU or GPU to accelerate your workload. For more information, see the guide to finding and using VM images on AI Hub.
Understanding public and private assets on AI Hub
Assets on AI Hub are collected in two scopes: public assets and private assets.
- Assets in the public scope are available to all AI Hub users. The collection of public assets contains AI resources that Google has published for general use.
- The private scope contains AI components that you have uploaded and assets that have been shared with you.
Sharing AI assets
Assets that you upload to AI Hub are not shared by default. You can choose to share these assets with individuals, groups, or with your entire organization. To learn more about sharing assets with your colleagues, see the guide to sharing assets on AI Hub.
Comparing AI Hub to TensorFlow Hub
TensorFlow Hub provides a library of TensorFlow modules that you can use to speed up the process of training your model. On the AI Hub, you can explore and use a variety of AI asset categories.