Baidu debuts updated AI framework and R&D initiative

At Wave Summit, Baidu’s bi-annual deep learning conference, the company announced version 2.1 of PaddlePaddle, its framework for AI and machine learning model development. Among the highlights are a large-scale graph query engine; four pretrained models; and PaddleFlow, a cloud-based suite of machine learning developer tools that include APIs and a software development kit (SDK). Baidu also unveiled what it’s calling the Age of Discovery, a 1.5 billion RMB (~$235 million) grant program that will invest over the next three years in AI education, research, and entrepreneurship.

At Wave Summit, Baidu CTO Haifeng Wang outlined the top AI trends from the company’s perspective. Deep learning with knowledge graphs has significantly improved the performance and interpretability of models, he said, while multimodal semantic understanding across language, speech, and vision has become achievable through graphs and language semantics. Moreover, Wang noted, deep learning platforms are coordinating closely with hardware and software to meet various development needs, including computing power, power consumption, and latency.

To this end, PaddlePaddle 2.1 introduces optimization of automatic mixed precision, which can speed up the training of models — including Google’s BERT — by up to 3 times. New APIs reduce memory usage and further improve training speeds, as well as adding support for data preprocessing, GPU-based computation, mixed-precision training, and model sharing.

Also in tow with PaddlePaddle 2.1 are four new language models built from Baidu’s ERNIE. ERNIE, which Baidu developed and open-sourced in 2019, learns pretrained natural language tasks through multitask learning, where multiple learning tasks are solved at the same time by exploiting commonalities and differences between them. Beyond this, PaddlePaddle 2.1 brings an optimized pruning compression technology called PaddleSlim, as well as LiteKit, a toolkit for mobile developers that aims to reduce the development costs of edge AI.

PaddlePaddle Enterprise and Age of Discovery

PaddlePaddle Enterprise, Baidu’s business-oriented set of machine learning tools, gained a new service this month in PaddleFlow. PaddleFlow is a cloud platform that provides capabilities for developers to build AI systems, including resources management and scheduling, task execution, and service deployment via developer APIs, a command-line client, and an SDK.

In related news, Baidu says that as a part of its new Age of Discovery initiative, the company will invest RMB 500 million ($78 million) in capital and resources to support 500 academic institutions and train 5,000 AI tutors and 500,000 students with AI expertise by 2024. Baidu also plans to pour RMB 1 billion ($156 million) into 100,000 businesses for “intelligent transformation” and AI talent training.

Laments over the AI talent shortage have also become a familiar enterprise refrain. O’Reilly’s 2021 AI Adoption in the Enterprise paper found that a lack of skilled people and difficulty hiring topped the list of challenges in AI, with 19% of respondents citing this as a “significant” barrier. In 2018, Element AI estimated that of the 22,000 Ph.D.-educated researchers working on AI development and research globally, only 25% are “well-versed enough in the technology to work with teams to take it from research to application.”

“PaddlePaddle researchers and developers will collaborate with the open source community to build a deep learning open source ecosystem and break the boundaries of AI technology,” Baidu said in a press release. “With the permeation of AI across various industries, it is critical for platforms to keep lowering their threshold to accelerate intelligent transformation.”