HUAWEI CLOUD ModelArts2.0: Revolutionizing AI Development
Sep 20, 2019 GMT+08:00
[Shanghai, China, September 20, 2019] At HUAWEI CONNECT 2019, Jack Jia, General Manager of HUAWEI CLOUD EI Service Product Dept, announced the release of a new one-stop AI development and management platform, ModelArts2.0.
HUAWEI CLOUD officially released ModelArts in 2018. So far, ModelArts has served more than 40,000 developers and is widely used in industries such as construction, Internet, and healthcare.
Jack Jia announced that ModelArts2.0 has made AI development simpler and more automatic. ModelArts2.0 is a massive leap forward in terms of data preparation, algorithm development, model training, model management, and model inference. ModelArts2.0 includes more than 10 new features and services, covering the entire life cycle of an AI model.
▶In terms of data processing, ModelArts2.0 automatically filters out invalid data for training models using the intelligent data filtering feature. For example, defocused or overexposed images will be filtered out for labeling in a visual analysis scenario. In business scenarios, other items will be filtered out.
By upgrading conventional active learning, hybrid labeling technology was initiated and can improve labeling efficiency by at least five fold.
In addition to automatic labeling, ModelArts2.0 will bring data evaluation and diagnosis functions online. Currently, ModelArts2.0 provides automatic feature extraction and visualization capabilities based on about 20 features, such as sample quality features, image properties, and labeling features. These features help developers identify deep, underlying relationships between data points, which can then be used for data optimization.
▶In terms of training, unique multi-element search technology integrates automatic data enhancement, automatic hyperparameter search, neural architecture search, and other technologies to help users quickly build AI models that outperform the average precision level for the industry.
The open ModelArts SDKs can be integrated into third-party development environments, for example, PyCharm. It streamlines development pipelines for enterprise-class applications, enables developers to flexibly take advantage of powerful cloud computing capabilities on demand. It supports collaborative online and offline development.
For senior algorithm engineers and data scientists who want to focus more on data modeling and problem solving, the Huawei AI computing framework, MindSpore, provides automatic parallel processing capabilities and powerful automatic differentiation capabilities.
▶In terms of model management, it is both very important and very difficult for algorithm developers to evaluate and optimize model performance. ModelArts2.0 provides extensive model evaluation APIs and visualization capabilities, and even offers model diagnosis suggestions for AI developers based on the evaluations. ModelArts2.0 helps developers optimize their models.
▶In terms of model inference, ModelArts2.0 is the first AI development and management platform in the industry to provide automatic hard example detection. It uses online learning to streamline the running and development of AI models. The models can learn and evolve online along with data increases and changes. ModelArts2.0 provides true enterprise-class, evolvable AI. This function has been tested on HUAWEI CLOUD Octopus.
Jack Jia said, "In 2019, Huawei completed the full stack AI capability integration from chips to software and HUAWEI CLOUD AI has been significantly upgraded. Through technical innovation, we hope to deliver a new AI development experience. We hope to significantly lower barriers to entry, to make it easier for those new to AI to get involved, and easier for existing software developers to rapidly master AI techniques. We want to significantly improve the efficiency of professional AI developers and to free up algorithm engineers and data scientists to focus on basic research into the core algorithms they use and to innovate new ones. We want to unleash the infinite potential that lays hidden within."