Build every part of machine learning
All powered by Google technology and tools
Tensor Projects
Provides developers with a complete set of
Outstanding and synergistic technologies and platforms
Helping developers unlock the future of machine learning
Follow in the footsteps of Google Developers Conference 2022
Explore the infinite possibilities of machine learning together!
Tensor Projects: Building a Comprehensive Open Source Product Ecosystem for Machine Learning
Google is committed to building a powerful ecosystem of machine learning tools that enable developers to turn ideas into products on all hardware and device types. With this goal in mind, the product team has developed several useful tools for both field-leading research and global deployment. These outstanding and synergistic technologies and platforms are collectively referred to as Tensor Projects.
Pillar One: Data Injection and Preprocessing
High-quality data helps build better machine learning models. Tensor Projects provides a variety of open source tools and resources to help developers acquire, label, preprocess and extract high-quality data to meet the requirements of high-quality model building .
- TensorFlow DataSets are a collection of ready-made datasets that can be used with TensorFlow or other Python machine learning frameworks such as Jax. Helps developers quickly build and validate machine learning model prototypes without having to spend a lot of time manually collecting and labeling data.
- Keras makes data preprocessing easier. Developers can use these features either outside the model (ie in the tf.data pipeline, suitable for training) or inside the model (suitable for inference).
Pillar 2: Model Architecture Definition and Training
Google offers developers a variety of options to help define and train machine learning models.
- JAX is a framework optimized for hardware accelerators, which can help developers delve deeper into the mathematical operations of machine learning and promote the development of machine learning research. For example, DeepMind used JAX to develop AlphaFold to solve protein folding problems to accurately predict protein structures.
- For developers who don't need to go deep into the math, TensorFlow makes it easy to create models. For developers developing mobile or embedded applications, TensorFlow Lite Model Maker can help solve many complex tasks in the process of creating models, such as data processing, training, evaluation, optimization, transformation, and more. The Model Maker and Task libraries now support the Searcher API for end-to-end large-scale neighbor search.
Pillar Three: Model Deployment
The trained model can be deployed to different scenarios as needed, such as cloud, web, browser, mobile, embedded platform, and even microcontroller.
- TensorFlow Extended (TFX) is an end-to-end platform for deploying production machine learning pipelines. With TFX, developers can easily build machine learning processes and infrastructure, and flexibly deploy machine learning models into production environments.
- As a very popular component of TFX, TensorFlow Serving can deploy models for multiple platforms and perform remote inference after machine learning model training is complete.
Pillar Four: Monitoring and Maintenance
The introduction of new data, bug fixes, performance improvements, and other elements require that we must continuously train and deploy models. This process of continuous learning and improvement as the model is used is called MLOps and is the fourth pillar. To this end, Google provides a series of tools to help developers more easily realize the continuous deployment of models, and TFX is one of the important tools.
MediaPipe: Simplifying on-device machine learning
MediaPipe provides developers with a customizable on-device machine learning solution for everyone. Complex pipelines can be encapsulated into MediaPipe Tasks, while making custom models easy for everyone with MediaPipe Model Maker. Provides customizable, high-performance on-device machine learning solutions through a low-code API. In addition, code-free graphical tools are coming soon. More and more powerful solutions are coming soon.
TensorFlowLite in Google Play Services: Facilitating App Publishing
TensorFlow Lite, built into Google Play Services, recently launched a beta version.
It can help developers greatly reduce the size of the application, and let users use the latest version in time through the background update function without redeploying the application.
"Introduction to TensorFlow - Deployment" login
The official TensorFlow team has launched the special course "Introduction to TensorFlow - Deployment". The course is based on "Machine Learning Practice: Model Construction and Application", designed and produced by teachers from many domestic colleges and universities and Google certified developer experts. The course will guide developers to quickly get started in the field of deployment, quickly master how to deploy models in Android and iOS projects, browsers using JavaScript, and service scenarios through the cloud, and prepare for practical use.
Welcome to leave a message in the comment area
Share your thoughts and expectations about Tensor Projects
Keep following Google Developers
Learn more about Google's technical progress and product updates
Explore the infinite splendor of "shared code future"!
**粗体** _斜体_ [链接](http://example.com) `代码` - 列表 > 引用
。你还可以使用@
来通知其他用户。