Machine Learning Flaw?

Musing this, is there really a flaw here? Hmm.  OK, you need enough data to fulfill the  accuracy needed.   So a reasonable caution.

Widely Used Machine Learning Method Doesn't Work as Claimed
UC Santa Cruz Newscenter
Tim Stephens
March 16, 2020

A study by researchers at the University of California, Santa Cruz (UCSC), Google, and Stanford University found fundamental flaws in a widely used machine learning (ML) technique for modeling complex networks. The researchers said low-dimensional embeddings have drawbacks, and mathematically showed that significant structural aspects of networks are lost in the embedding process. UCSC's C. Seshadhri warned that any embedding technique yielding a small list of numbers will basically fail because a low-dimensional geometry is insufficiently expressive for social networks and other complex networks. Seshadhri said the research shows the need to check the validity of underlying ML assumptions, because "in this day and age when machine learning is getting more and more complicated, it's important to have some understanding of what can and cannot be done."