Baidu debuts latest version of AI platform Baidu Brain
The company says its Baidu Brain, which has over 270 AI capabilities, is used internally as well as being open to third-party developers, having been involved in 310,000 developer models.
The platform takes a layered approach to AI, with a foundation layer supported by additional layers of algorithms supporting features such as VR, voice technology and computer vision, as well as a security layer.
In , the company said: “In computer vision, state-of-the-art efficiency is becoming a new trend to meet an increasing demand of edge-based computer vision applications. Baidu Brain 6.0 open sourced a series of lightweight vision models including the model that topped the ‘Real-time Image Classification Using Pixel 4 CPU’ competition at CVPR 2020 and a 3.1M OCR model.”
The foundation layer involves the company’s deep learning (the subset of machine learning that takes inspiration from the brain) platform PaddlePaddle, which has been updated with 140 new APIs for better connection with third-party feature sets.
The company’s offering also includes Kunlun, a “cloud-to-edge AI chip” specifically built for AI scenarios, that the company produces in house and claims outperforms Nvidia’s T4 GPU/
“Artificial intelligence is the core technology of the fourth-generation industrial revolution,” said Baidu CTO Haifeng Wang. “Baidu Brain can enable all industries to apply AI technology more efficiently, while accelerating the process of industrial intelligence.”
Other announcements made at the event included its autonomous driving platform Apollo, which has completed over six million kilometres of on-road testing, and now is capable of using 5G communications technology to replace in-vehicle safety drivers.