How Huawei’s ‘Genius Teen’ Scaled AutoML to Millions of Smartphones

Huawei highlighted a young researcher named Zhong Zhao (钟钊) who joined the company in 2019 with a 2 million RMB salary and, in less than a year, led a team to deploy AutoML‑driven algorithms on millions of Mate and P series smartphones.

Technical Challenge: Pixel‑Level Image Processing

Zhong focused on the difficult problem of balancing algorithm accuracy and model size for pixel‑level image processing, enabling features such as super‑resolution and space‑enhancement to run efficiently on mobile devices.

Applying AutoML to Pixel Algorithms

While AutoML had been widely used for image classification and object detection, applying it to pixel‑level tasks was unprecedented. Zhong’s team built an end‑to‑end pixel‑level AutoML pipeline, extending Huawei Noah’s Lab VEGA framework (including CARS and ESR‑EA) with a dynamic convolution‑kernel method that adapts to image content, reducing computation by 37‑71 % without sacrificing accuracy.

Results and Impact

The technology reportedly reduces the complexity of video‑photography prototype algorithms by two orders of magnitude and is already shipped in several new phone models, with plans for broader rollout across future devices.

Additional Contributions

Zhong also addressed redundancy in lightweight mobile‑vision models (e.g., ShuffleNet, MobileNetV3) by proposing dynamic kernel generation, and he developed an adversarial automatic data‑augmentation technique published at ICLR 2020.

Biography and Academic Background

Born in 1991 to a computer‑science family, Zhong earned a software‑engineering degree from Huazhong University of Science and Technology, won a provincial award in the national college mathematics modeling contest, and later worked at the Chinese Academy of Sciences under Deputy Director Liu Chenglin.

During a 2018 internship at SenseTime, his first‑author paper on a block‑generation method for high‑performance neural networks was accepted as a CVPR Oral, a rare achievement for Chinese researchers.

His Ph.D. dissertation focused on “Deep Neural Network Architecture: From Manual Design to Automated Learning,” and he has published in IEEE TPAMI, ICLR, ICCV, and NeurIPS.

Open‑Source and Publication Links

The VEGA AutoML repository is publicly available at https://github.com/huawei-noah/vega, and his Google Scholar profile lists over 400 citations.

Public Perception

Despite some external criticism of the high salary, Huawei’s official release frames the achievements as validation of both Zhong’s expertise and the company’s confidence in large‑scale AutoML for mobile AI.