Midea, one of the world’s largest manufacturers of household appliances, has launched an ambitious artificial intelligence project through which the company is looking to transform its business.
The company says the “Midea Brain” project will accelerate Midea’s speed of innovation and product and application development in the AI field.
In the last two years, artificial intelligence especially deep learning, has emerged as a major technology trend.
However, powerful parallel computing power has become a necessity in the development and training of highly accurate deep learning models.
Just one week ahead of Midea Group’s 2018 Strategy Conference, the Midea Global AI Team officially announced the Midea Brain, a peta-scale, high-density, distributed, heterogeneous deep learning platform, showcasing Midea’s determination and investment in its AI-driven transformation.
The key part of Midea Brain is its high density, high efficiency, heterogeneous computing platform designed for deep learning training.
Custom designed and built by Midea’s Global AI Team, especially the AI team in Silicon Valley, the coherent hardware and software platform consists of several hundred Nvidia Tesla P100 GPUs.
It supports Nvidia Volta series GPU, and is capable of Peta FLOPS peek computing speed.
Compared to the commonly used parameter server framework in a GPU cluster, Midea’s AI team proposed and built what they describe as an “innovative Job Server framework”, which significantly reduces the GPU waiting time, and dramatically increases the performance of the GPU clusters.
Midea Brain platform software is designed and developed by the Midea AI team, and is said to be compatible with AI models trained by other popular AI frameworks, such as TensorFlow and Caffé.
This software utilizes a distributed framework, supports both large-scale GPU clusters and a single GPU node.
Therefore, it can easily scale and maximize efficiency for AI projects with varying computational needs.
This platform enables Midea to independently develop and customize AI models and algorithms for its products.
It also significantly increases Midea’s product development speed, efficiency and reduces the dependence on third-party AI software.
Dr Wang Dongyan, Midea Group’s global AI team leader, vice president and general manager of the Silicon Valley Emerging Technology Center and Shenzhen Artificial Intelligence Institute, says: “Computing power is absolutely essential for AI, especially deep learning.
“Assuming the same model and time constraint, the stronger the computing capability, the more learning experience will be accumulated, and the more model iteration we can achieve.
“The development of Midea’s peta-scale, heterogeneous deep learning platform greatly expedites the training speed of Midea’s AI deep learning models.
“Furthermore, it boosts Midea’s speed of innovation and product and application development in the AI field.
“Based on the Midea Brain deep learning platform, our Global AI team has successfully developed, trained, and promptly iterated multiple AI products in 2017, including the Midea smart oven food recognition, the AI cooking robot, Chinese dish recognition, and industry AI.”
He adds: “Most traditional enterprises don’t have the abundant GPU and CPU computing resources that large Internet companies have in their data centers.
“With the advantages of high performance/cost ratio and high efficiency in both hardware and software, Midea Brain can provide a great reference for companies that want to adopt AI as the engine of transformation.
“Midea Brain’s compact heterogeneous deep learning platform satisfies the needs of large-scale AI computing ability, without breaking the bank.”
As a technology group focusing on consumer appliances, HVAC systems, robotics and industrial automation systems, and smart supply chain or logistics, Midea believes that AI is the core competency to drive the group’s transformation.
It says it is “moving from a traditional Chinese home appliance manufacturer to a global high-tech group”, with independent, innovative smart home and manufacturing systems.
Midea Brain, the proprietary large-scale, heterogeneous, deep learning platform, will form a critical foundation for Midea in developing many AI applications and products, and will help build the “smart home + intelligent manufacturing” ecosystem.
The company adds that, in the near future, with its “advantages in AI platform and technology, and the backing of a strong global AI team, Midea will accelerate its AI innovation, and launch AI products that will benefit the consumers”.