The Origin Of The Specious: Which Algorithm Is Best?
by Joe Mandese, Feb 21, 2013, 6:29 PM
One of the reasons I wanted to edit RTM Daily and write RTBlog is to do the deep dive into all this geeky technology that’s transforming our industry (and society), but even I wasn’t prepared for the conversation that took place Wednesday afternoon on the West Side of Manhattan where Rubicon Project held its New York Summit updating the industry on some new bells and whistles and, well, algorithms. In fact, those mathematical programs were the real stars of the event, even though Rubicon brought some pretty big industry execs on stage, and surrounded the audience with some sci-fi looking server stacks that would make Hal 9000 rust with envy.
My main takeaway: Not all algorithms are created equal, and not all of them optimize the same things. For example, there are “decider algorithms” that process trillions of permutations extremely fast (a lot faster than we can think) to sort and parse all the inventory data that’s possible to buy in the RTB marketplace. Then there are “informer algorithms” that make the actual recommendations that traders can decide on.
But the most interesting conversation about algorithms took place during a panel moderated by Rubicon Chief Scientist Neal Richter, who was surprisingly plain-spoken and understandable, acting more as a translator at times than a moderator.
The best question he posed was when he asked his panel to name their “favorite algorithm.”
Tarun Bhatia, principal research scientist at Yahoo, said his favorite was the “inventory management algorithm” he built for Yahoo, which is the heart of the system that classifies and allocates all of the premium inventory in Yahoo’s pool.
Nicholas Kruchten, head of product engineering at Datacratic, said his favorite was a “proportion algorithm” which he likened to a hot and cold water valve you use to get the temperature of the water in your shower just right.
Sandro Catanzaro, vice president-analytics and innovation at DataXu said his favorite was an algorithm capable of parsing the sentiment a consumer has for a brand in order to get beyond things like “clicks and conversions.”
But my personal favorite was the one that Rubicon’s Richter also described as his personal favorite: An “evolutionary algorithm.”
“You just change something and you see what happens,” he explained. “This is essentially evolution... a very simple algorithm, but very effective over time.”
Hmmm, I thought Charles Darwin wrote that code.