Baidu’s open-sourced deep-learning platform paddles itself to China’s No.1 and world’s Top 3
From browsing short videos to launching rockets into space, our lives are intertwined with AI technology. On average, it takes 3-6 months to put an artificial intelligence application into use. It is usually a complicated process to develop AI solutions and customize them according to different scenarios.
“When faced with challenges, technology will always evolve as demands grow,” said Dr. Ma Yanjun, president of Baidu’s AI technology ecosystem, at an AI-themed conference hosted by Baidu on March, 31. Ma said in order to advance AI technology and create solutions tailored to various scenarios in various industries, “open-sourced platforms are born and can help simplify complicated issues.
According to China’s 14th Five-Year Plan, “deep learning framework” was included in the field of “new generation of artificial intelligence” and became a key frontier innovation technology supported by the state. As Ma put it, deep learning platforms are crucial to building an AI ecosystem, since “the deep learning framework middle part of the artificial intelligence technology system, connecting basic infrastructure technology such as chips to the final application.”
When an open-sourced deep learning platform emerges, it proves to users that it’s trustworthy, fosters the growth of the entire AI industry by allowing other developers to build on existing foundations. Eventually, as developers and users increase, the open-sourced platform itself becomes an industry standard.
Founded in 2016, PaddlePaddle, Baidu’s own open-sourced deep-learning platform has accumulated 4.06 million developers, created 476,000 AI models, and served 157,000 enterprises and institutions, covering industries such as agriculture, healthcare, urban development, transportation, finance, and others. PaddlePaddle supports 31 kinds of chips from 22 manufacturers, including Intel, NVIDIA, Baidu, and Huawei, covering all the mainstream chips in China and abroad.
As Ma explains, PaddlePaddle’s name is abbreviated from the pedantic term Parallel Distributed Deep Learning. Also, it’s PaddlePaddle, not Paddle since “two paddles will make the boat go faster”.
Currently, PaddlePaddle has become the world’s top 3 deep learning platforms and counts Google’s TensorFlow and Meta’s Pytorch as its competitors. PaddlePaddle also ranks as the biggest deep learning platform in China, as it has the most market share, according to an IDC report.
In the wake of the COVID-19 outbreak, PaddlePaddle was used to develop a pneumonia CT imaging analysis system. The system dramatically decreased the time needed to measure and diagnose pneumonia from 2 hours or more to tens of seconds.
In addition, PaddlePaddle was also used to develop an image recognition system for the IFAW (International Fund for Animal Welfare), helping law enforcement to trace illegal animal transactions with 75% accuracy.
“As China’s deep learning frameworks become more accessible due to open-source platforms, they will have more applications in the future. The cost and application threshold will also be further reduced. Meanwhile, deep learning frameworks will be integrated and innovated with more frontier industries such as scientific computing, quantum computing, and life science,” said Ma.
Image Credit: Zijing Fu/PingWest