Baidu’s homegrown deep learning framework PaddlePaddle will empower Huawei’s Kirin smartphone chips, the company announced at the Baidu Create 2019 AI Developer Conference that kicked off today in Beijing. The long-awaited partnership between Baidu and Huawei is expected to expand the AI capability on Huawei’s smartphones, which now account for 19 percent of the global smartphone market.
PaddlePaddle was introduced and open-sourced by Baidu in 2016 as an easy-to-use, efficient, flexible and scalable deep learning platform — think of it as Baidu’s “TensorFlow.” PaddlePaddle is used internally by Baidu engineers for research and has helped with the development of many Baidu products. Company CTO Haifeng Wang says PaddlePaddle supports over 70 popular machine learning models.
Baidu and Huawei has been working closely for years, and the partnership announced today focuses on three main hardware and software objectives:
- PaddlePaddle will be fully integrated with HiAI, the mobile computing architecture Huawei introduced for AI acceleration on mobile chips. The goal is to maximize the potential of chip hardware and deliver more powerful compute power for edge AI.
- Baidu and Huawei will jointly optimize classic machine learning models to enable devices equipped with Kirin chips to run AI applications with greater speed and efficiency.
- Leveraging PaddlePaddle’s performance and functionality, Baidu will help Huawei improve Kirin’s compute power and accelerate the creation of next-generation chips.
Other announcements at Baidu Create 2019 include:
- New chip: Baidu introduced a new chip to enable far-field voice interaction. Honghu (鸿鹄) is an automotive-grade processor featuring two HiFi 4 DSPs and 100mW power consumption. This new chip can support high-precision voice awakening with ultra-low false alarms and offline voice recognition, making it suitable for scenarios such as smart home and in-car interaction.
- Baidu Brain 5.0: Baidu Brain is the baseline for Baidu AI services ranging from image recognition to natural language understanding. Announced today, Baidu Brain 5.0 adds Baidu’s latest self-developed algorithms and models such as ERNIE, a new knowledge integration language representation model which outperforms Google’s state-of-the-art BERT in Chinese language tasks; and SMLTA (streaming multi-layer truncated attention model) for online automatic speech recognition.
- Three new speakers: Baidu launched three new Xiaodu smart speakers — “King Kong” , “Play” and the “Xiaodu at Home 1C 4G” battery version, all priced at CN￥199.
- DuerOS 5.0: Baidu updated its voice-based smart system DuerOS, adding a new ‘full-duplex’ ability that allows Xiaodu to respond without a wake word, or continued conversation, and a feature that enables the system to decide whether to reply and execute tasks or just listen. The DuerOS voice assistant install base has surpassed 400 million and monthly voice queries on DuerOS exceeded 3.6 billion.
- Apollo 5.0: Baidu upgraded its open-source autonomous driving Apollo Open Platform and Apollo Enterprise. The company also announced its autonomous vehicle fleet has now driven more then two million kilometers (over 1.24 million miles) in 13 cities.
- Partnership with Intel: Baidu and Intel announced they will partner on Intel’s new Nervana Neural Network Processor for training.
- Partnership with Geely: Geely cars, beginning with the Geely Boyue Pro, will be equipped with the GKUI19 connected vehicle solution powered by Baidu’s DuerOS for Apollo.
There was an odd disturbance during CEO Robin Li’s keynote speech, when an audience member charged the stage and splashed Li with a bottle of water. The intervention surprised the Baidu CEO, who recomposed before calmly remarking “There are many difficulties on the road to achieve AI, but they will not impact Baidu’s determination.” Li’s words were welcomed by the audience with enthusiastic clapping and chants.
Journalist: Tony Peng | Editor: Michael Sarazen