ChinAI #176: AI Frameworks (part 2)
Greetings from a world where…
the Hawkeyes rule the Big Ten
…As always, the searchable archive of all past issues is here. Please please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay support access for all AND compensation for awesome ChinAI contributors).
Feature Translation: AI Frameworks Development White Paper
Context: Last week, we covered a white paper (original Chinese), published by the China Academy of Information and Communications Technology (CAICT), on the development of “AI Frameworks” — standard interfaces, libraries, and toolkits for designing, training, and verifying AI algorithms. The first half of the white paper provided an overview of the global AI framework landscape, revealing not only the dominance of TensorFlow and PyTorch but also the growth of Baidu’s Paddle Paddle and Huawei’s Mindspore. This week, we finish the rest of the white paper.
Key Takeaways:
Academic and industrial applications of AI frameworks in China:
Huawei/Pengcheng Laboratory released PanGu, a GPT-3-like language model, in May 2021, which was developed on MindSpore’s framework (see ChinAI #141 for more background)
Huawei's factory in Songshan has also introduced MindSpore and AI quality inspection algorithms to improve the defect detection in printed circuit boards.
Linking Med has launched a pneumonia screening AI system based on the Paddle Paddle platform, which has been used at a hospital in Chenzhou, Hunan province.
There’s a lot of impressive statistics in this section of the white paper, but it reads a lot like PR-type language, so what’s important is to look for the types of applications rather than the numbers.
Promotion channels for Chinese AI frameworks:
Again, all the examples surround Huawei’s MindSpore and Baidu’s Paddle Paddle. They are trying to build open source communities and incentivizing researchers to use their frameworks.
Huawei: Mindspore developer support program that gives developers preferential cloud service resources, support for Ministry of Education’s industry-university cooperative education project, partnership with Chinese Association of AI on an academic award fund that has invested $16 million toward 120 Mindspore-linked projects.
Baidu: 150 Paddle Paddle City/University Pilot Groups, 132 communities are organizing Paddle Paddle activities in Chinese cities and universities, partnered with China Computer Federation on a “Pinecone Fund” that promotes the application of AI frameworks in scientific research.
Future trends in AI frameworks:
“Hyperscale AI” — using the example of PanGu, the white paper highlights the importance of AI frameworks in helping with efficiently training large models by intelligently utilizing computer power resources
Secure and trustworthy AI — white paper lists some platforms that evaluate the interpretability of AI models. PyTorch has Captum; TensorFlow has TF-explain. In China, there’s Mindspore’s XAI and Paddle Paddle’s Interpret DL, as well as the Chongming platform of OpenIntelligence (OpenI) and the RealAI platform.
AI engineering-ization (工程化): a top strategic technology trend, according to Gartner, for both the past two years. This tackles nitty gritty issues involving how to compress AI models and develop lightweight versions of models to run on various terminal devices.
FULL TRANSLATION: CAICT AI Frameworks Development White Paper
ChinAI Links (Four to Forward)
Should-read: Green AI and large AI models (in Chinese)
Published in keyanquan, a platform under the Chinese-language version of Scientific American, this article mulls the environmental cost of large AI models. Includes a lot of discussions about the “East-West division of computing labor” project (based on transferring data centers to Western provinces where renewable energy is more competitive), which I flagged in ChinAI #159. H/t to Wendy Liu for recommending this article to me.
Should-read: Open source creates value, but how do you measure it?
Last week’s issue cited a bunch of indicators about AI frameworks derived from Github statistics. Peter Cihon’s Github blog post argues that open source development has been neglected as a metric of innovation, and highlights a few areas to explore further.
Should-read: Europe is in Danger of Using the Wrong Definition of AI
Joanna J. Bryson, for Wired, emphasizes the importance of using the right definition of AI in debates over the European Union’s regulations on AI. Bryson notes that some moves to restrict the definition of AI could let companies avoid oversight.
Should-read: The innocent have paid a high price for the Post Office scandal. The guilty have not
Bryson’s article also linked me to a fascinating historical story (recounted by Marina Hyde for The Guardian):
In the spirit of rearranging our heads once more, let’s do the brief summary: between 2000 and 2014, 736 subpostmasters and postmistresses were prosecuted of theft, fraud and false accounting in the branches of the Post Office they ran. Their lives – and the lives of thousands of others – were torn apart. They were financially ruined, put out of work, locally shunned, driven into poor health and addiction, saw their marriages destroyed. Some – from a 19-year-old woman to mothers of young children to all manner of others – were imprisoned for many months. At least 33 victims of the scandal are now dead; at least four reportedly took their own lives. But … they had done nothing wrong. They had done nothing wrong. The blame in fact lay with Horizon, a faulty computer system designed by Fujitsu and imposed on their branches by Post Office management.
Thank you for reading and engaging.
These are Jeff Ding's (sometimes) weekly translations of Chinese-language musings on AI and related topics. Jeff is a postdoctoral fellow at Stanford's Center for International Security and Cooperation, sponsored by Stanford's Institute for Human-Centered Artificial Intelligence.
Check out the archive of all past issues here & please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay for a subscription will support access for all).
Any suggestions or feedback? Let me know at chinainewsletter@gmail.com or on Twitter at @jjding99