2019 Q2 Release
Originally Posted on July 25, 2019
PCI-China's 2019 Q2 updates;
Facebook's open-source spree; and
Authorship attribution with machine learning.
Dear Human Readers,
Apparently, American and Chinese negotiators are set to resume trade talks—again—next week. While the two sides may have a lot to catch up on, our PCI updates are relatively more straightforward.
PCI-China’s 2019 Q2 updates. As mentioned in last month’s pre-release, the PCI project is maintaining its earlier assessment that China is turning more hard-line in its (domestic) economic policy and foreign relations. This has been the case before and after the trade war started, which means President Trump’s pressure campaign against China hasn’t shaken the latter. Given Washington’s increasingly bipartisan consensus over confronting China, one shouldn’t be surprised to see more uncertainties ahead.
The chart below shows the PCI-China index up to the second quarter this year, with the two recent spikes indicating China’s turns to (more) aggression. In a blog post about the Q2 updates, Weifeng Zhong and Julian TszKin Chan also discussed the “Soviet root” of the PCI algorithm; the way Vladimir Levin and Josef Stalin shaped the USSR nearly a century ago turns out to have much to say about the ongoing US-China trade war.
Figure: PCI for China, 1951 Q1 to 2019 Q2
Note: The PCI-China predicts if and when the Chinese government will change its policy priorities. A spike in the indicator signals a policy change, while a vertical bar marks the occurrence of a policy change labeled by the event.
Facebook is on fire. What we meant is not the regulatory and legal troubles the company has gotten into, but the spree of artificial intelligence technologies it has made publicly available in recent months. Among others, the tech giant open-sourced its reinforcement learning platform Horizon and natural language processing framework PyText late last year, image-processing library Spectrum in January, and, just this month, its controversial deep learning recommendation models DLRM.
Facebook’s open-source efforts are making it easier for everyone else in the community to innovate. But even for the tech empire’s self interest, the act of open-sourcing, in itself, is good business. Reasonable people can disagree about how much they hate Facebook, but the tech giant’s willingness to open-source its secret sauce is much welcomed news.
What We're Reading
Authorship attribution with machine learning. Many of The Beatles’ most famous songs were credited to “Lennon-McCartney”—the duo’s songwriting partnership—even if a song was mostly written by just one of the two. This arrangement has made it difficult to tell who wrote what.
Just this month, a machine learning algorithm developed by Mark Glickman, Jason Brown, and Ryan Song—a group of Harvard University and Dalhousie University researchers—took up the authorship attribution challenge. They cleverly turned music into data by coding each song according to its musical features, such as melodic notes and chords. How well did the algorithm do, you ask? Well, Sir Paul declined to comment, unfortunately.
The researchers’ endeavor brings back memories of the ground-breaking work by statisticians Frederick Mosteller and David Wallace, who in their 1963 paper used Bayesian models to infer that James Madison, rather than Alexander Hamilton, wrote all 12 of the disputed Federalist Papers. Too bad the Founding Fathers can’t possibly comment either.
Edited by Weifeng Zhong and Julian TszKin Chan
If you were forwarded this email and like what you read, please sign up here to stay in the loop. If you’re already a loyal reader, please help us grow by forwarding this to a friend or colleague.