Core Functions of AI Development
Up to 80% reduction in manual data processing costs
ModelArts includes nine labeling tools to manage four types of datasets including text, images, audio, and video. It also offers auto labeling and team labeling, for more efficient dataset labeling. Data processing capabilities, such as data cleansing, enhancement, and verification, backed up by flexible, visualized management of dataset versions are also available. Import and export your datasets with ease as you develop and train your models.
Combine cloud resources with the development toolchain for more efficient AI development
ModelArts provides out-of-the-box cloud-native notebook instances, allowing you to flexibly use cloud resources for accelerated AI development and debugging. This mode is tailored for AI beginners.
You can also remotely access cloud resources through a local IDE (PyCharm) or VS Code plug-ins to customize your development environment. This mode is tailored for AI professionals.
Algorithm suites, pre-trained models, and AI assets from multiple communities or Huawei are provided through algorithm engineering. This further empowers the algorithm development of ModelArts.
Train high-precision models faster
Powered by an EI-Backbone, ModelArts workflow excels at:
- Greatly reducing the cost of data labeling by training high-precision models using small volumes of data.
- Quickly improving the model precision by using the full-space network architecture search and automated hyper-parameter tuning.
- Significantly reducing training costs by employing pre-trained models to shorten the time required for deploying a trained model from weeks to minutes. All of these promote inclusive AI.
Manage all iterated and debugged models in a unified manner
AI model development and tuning require frequent iterations and debugging but changes in datasets, training code, or parameters could affect model quality. If the metadata of the development process cannot be managed in a unified manner, the best solution may fail to be reproduced. ModelArts allows you to import models generated with all training versions from training jobs, templates, container images, and OBS.
One-click deployment of models to the device, edge, and cloud
ModelArts models can be deployed as real-time, batch, or edge services. Real-time services process a large volume of highly-concurrent data, while batch services feature high-throughput capability of quickly processing data. Edge services feature the capability of completing inference locally in a highly flexible way.
Custom image function allows you to customize engines
ModelArts uses container technology at the bottom layer so you can create container images and run them on ModelArts. The custom image function supports command line parameters and environment variables in free-text format. The custom images are highly flexible and support the job boot requirements of any computing engine.
ModelArts-empowered developer ecosystem community
In this community, scientific research institutions, AI application developers, solution integrators, enterprises, and individual developers can share and purchase AI assets such as algorithms. This accelerates the development and implementation of AI assets and enables every participant in the AI development ecosystem to achieve business success.
See how ModelArts has enabled projects in science, environmental protection, and games
Rainforest Connection (RFCx) is a non-profit organization dedicated to protecting the world's rainforests. RFCx used Huawei Cloud AI services and ModelArts to build an intelligent model capable of detecting and decoding the spider monkey sounds, to learn about their habitat, of threats to their survival, and about their daily behaviors. This information helps the forest rangers keep the spider monkeys safe.
Center for Excellence in Brain Science and Intelligence Technology, the Chinese Academy of Sciences
ModelArts can automatically trace and reconstruct neurons with an accuracy and recall rate of up to 95%. Using parallel computing enabled by the ultra-large clusters of ModelArts, the total time required for morphological reconstruction of 100,000 neurons can be reduced from 125 person-years to just 10 days, and the cost of reconstructing a single neuron can be reduced to 1/77 the original cost. If the study were to be carried out on mice or macaques, the cost reduction and efficiency gains would be even more significant.
Dark Forest AI
Dark Forest, a hugely popular multiplayer shooting game created by Xishan Ju Game, used the ModelArts heterogeneous reinforcement learning platform to build intelligent, versatile robots with human-like gaming skills. The robots can pick up stars and props in the same way real players would, quickly beat competitors, and use strategies to win seesaw battles. Real players can choose to play with them for a brand-new and more enriching gaming experience.