Their approach, called algorithms with predictions, takes advantage of the insights machine learning tools can provide into the data that traditional algorithms handle. These tools have, in a real way, rejuvenated research into basic algorithms.
Machine learning and traditional algorithms are “two substantially different ways of computing, and algorithms with predictions is a way to bridge the two,” said Piotr Indyk, a computer scientist at the Massachusetts Institute of Technology. “It’s a way to combine these two quite different threads.”
Currently, computer scientists often design their algorithms to succeed under the most difficult scenario — one designed by an adversary trying to stump them. For example, imagine trying to check the safety of a website about computer viruses. The website may be benign, but it includes “computer virus” in the URL and page title. It’s confusing enough to trip up even sophisticated algorithms.
Indyk calls this a paranoid approach. “In real life,” he said, “inputs are not generally generated by adversaries.” Most of the websites employees visit, for example, aren’t as tricky as our hypothetical virus page, so they’ll be easier for an algorithm to classify. By ignoring the worst-case scenarios, researchers can design algorithms tailored to the situations they’ll likely encounter. For example, while databases currently treat all data equally, algorithms with predictions could lead to databases that structure their data storage based on their contents and uses.
Kinda reminds me of that scene in Friends where Monica vacuums her vacuum cleaner and says “if only there were a smaller one to clean this one”. But it’s ultimately humans who pull the strings of the meta-optimisation.