Huawei releases faster full-lifecycle AI development platform

Huawei has released faster full-lifecycle AI development platform.

  • Monday, 15th October 2018 Posted 6 years ago in by Phil Alsop

At HUAWEI CONNECT 2018, Huawei released its faster AI development platform ModelArts, vision AI application development platform HiLens, and quantum computing software cloud service platform HiQ.

 

AI development is inefficient and time-consuming, and AI resources, especially computing power, are both scarce and expensive. Through the newly released products, Huawei is endeavoring to accelerate AI industry development with developers.

 

ModelArts, A Faster and More Inclusive AI Development Platform

ModelArts is a faster and more inclusive one-stop AI development platform, involving data labeling and preparation, and model training, optimization, and deployment. It delivers end-to-end AI application development.

 

"ModelArts is a faster and more inclusive AI development platform than any other AI development platform out there," said Zheng Yelai, Vice President of Huawei and President of Huawei's Cloud BU, "We believe AI developers will appreciate how quickly it starts, completes training, and deploys models."


Data labeling and preparation is time-consuming in AI development, accounting for almost 50% of the time required. ModelArts has a built-in data governance framework for data labeling and preparation during AI development. The framework implements iterative training to reduce the volume of data that needs to be manually labeled, which improves the efficiency of data labeling and preparation by 100-fold.

 

In addition, ModelArts integrates multiple optimization technologies, especially cascade hybrid parallel to halve the duration of training required for a given model, dataset, and set of hardware resources.

 

AI model deployment is complex. For example, updated models for smart transportation need to be deployed on various types of cameras at a time, which takes a long time to complete. With ModelArts, training models can be pushed to devices, the edge, and the cloud with one click. Online and batch inference jobs are provided on the cloud to meet different application requirements, such as concurrent or distributed deployment.

 

Various AI technologies are also incorporated into ModelArts, including automated learning, model design, and parameter tuning to accelerate AI development.

 

In terms of managing the AI development lifecycle, ModelArts covers raw data collection, data labeling, training job creation, algorithm selection, model creation, and inference service creation. Visualized management of the entire lifecycle and automatic generation of model traceback diagrams allow users to manage and trace millions of datasets, models, and services. Resumable training and training result comparison are also supported.

 

ModelArts allows AI developers to share AI data, models, and APIs. Internally, ModelArts supports enterprises in sharing data and models, aiming to improve AI development efficiency, build AI development capabilities, and protect AI assets. Externally, Huawei is hoping to create an open ecosystem through ModelArts that will help AI developers monetize technologies and build influence and their own ecosystems.

 

HiLens, A Vision AI Application Development Platform

Huawei's second major release at HUAWEI CONNECT 2018 was HiLens.

 

HiLens comprises a vision AI application development platform as well as a visual device enhanced with AI capabilities. HiLens features Skill, a new concept of AI development. Skill consists of control code and models trained on ModelArts. HiLens is also compatible with models trained by other mainstream frameworks. Skills developed on HiLens can be deployed on any devices that integrate the Ascend AI chips.


The HiLens visual device is a smart camera that supports inference. Developers can use the HiLens device to develop vision applications, which can be deployed on devices and the cloud. The HiLens visual device integrates the Ascend 310 chip, which can process 100 frames per second and detect faces in milliseconds. In addition, the built-in lightweight containers minimize resource usage and network bandwidth usage, and can be quickly downloaded and started.

 

HiLens functions like a "smart eye" for AI applications. It can be widely used in scenarios like companion robots, drowsiness monitoring, hazard identification, and logistics. Huawei hopes developers can contribute their ideas and create more applications through HiLens.

 

HiQ, A Future-Ready Quantum Computing Simulator and Programming Framework

Huawei's exploration of the quantum field has culminated in its release of HiQ, a quantum computing software cloud service platform to provide developers with superior computing power and new algorithms.

 

Built with the quantum computing simulator and programming framework, HiQ features full-amplitude and single-amplitude simulators. Quantum error correction is integrated for the first time, realizing error correction in the simulation of tens of thousands of quantum bits. HiQ also incorporates a simplified hybrid classical-quantum programming and Block User Interface (BlockUI).

 

Based on the Huawei Cloud's powerful computing strength, HiQ can simulate at least 42-qubit circuits considering the entire amplitude, and at least 81-qubit circuits within a single amplitude. In addition, the platform can simulate a low depth circuit, and the qubit count can reach 169 within a single amplitude. Huawei's quantum computing software platform will be open to the public as an enabling platform for research and education.

 

Huawei hopes HiQ can help partners from academia and industry cope with challenges of quantum computing and cultivate quantum computing and quantum algorithm engineers.

 

Build the AI World with Developers

Through ModelArts, HiLens, and HiQ, Huawei Cloud is looking forward to exploring the intelligent world with developers, accelerating AI implementation in hundreds of industries, and providing inclusive AI platforms that are affordable, effective, and reliable.