Data Culture

This past week’s discussion of Big Data seems incomplete without mentioning machine learning. For the uninitiated, machine learning is a method of computing that “trains” algorithms to achieve a desired result using large amounts of data. While machine learning has been around since 1959, it has only recently come back into fashion as businesses are waking up to its potential in the age of Big Data.

The key ingredient in effective machine learning applications, from FaceBook to Baidu, is data. Lots of data. Because of their scale, these companies are able to gather trillions of data points for everything, including individuals’ emotions, shopping habits, facial features and much more. To a human, or even an army of humans, one trillion pictures of peoples’ faces would yield little utility. But, to an elite team of computer scientists, such pictures allow them to construct systems that can recognize identity and emotion more accurately, and at exponentially greater scale, than human beings.

The paradigmatic shift from silos of individual networks to aggregated data and computing resources known as cloud computing brings with it greater efficiency and more data. Joseph Sirosh, Microsoft’s VP of Machine Learning, was recently interviewed by the cloud and data experts at GigaOM. Sirosh explains that Microsoft is rapidly transitioning from an operating system provider to a cloud provider with expertise in big data and machine learning. He goes so far as to state that computing itself is less important than the data that it provides:

“I think you should even first ask, ‘How big is the world of data to computing itself?’” he said. “I would say that in the future, a huge part of the value being generated in the field of computing . . . is going to come from data, as opposed to storage and operating systems and basic infrastructure. It’s the data that is most valuable.”

Microsoft has made its billions by providing software and services. Pivoting the business model of a $360B company is a herculean task; it’s safe to say that Microsoft and every other major player in technology wouldn’t be chasing desperately after data science if it wasn’t a huge deal. Data is often described as the new oil of the 21st century – machine learning is the new refinery.

Continue reading Data is the Oxygen to Machine Learning’s Fire

During Tuesday’s class, Dr. Sample played a video clip advertising the new Walking Dead video game, which was relevant because of how the game kept track of individual decisions and compiled statistics aggregating all of its users’ choices. Forced to make heat-of-the-moment decisions—“Do I save Ben, or let him fall to his death?”—users were likely to choose options they would later second-guess. The aggregated stats gave context for those split-second decisions, either comforting those in the majority (“I felt bad for letting him die, but at least 79% of users did the same thing”) or compounding regret for the few (“Wow, that really was stupid to try and save him”). Generally, people find comfort in choosing with the majority, although there are certainly rogues out there who would intentionally take the less-beaten path.

What interests me is the difference between how people choose when given the stats as opposed to when it’s simply a blind choice—particularly in the context of a NCAA tournament bracket. Every March, thousands of people join online bracket-picking tournaments, often relying on picking the favorites in each first-round matchup (especially for games involving schools no one’s ever heard of). I imagine that, rather than blindly guessing, most people would lean toward the team that has, say, 71% of users picking them, rather than the underdog with 29% on their side. What intrigues me about this is the possibility of a snowball effect, where people disproportionately favor the team with 52% support, which pushes the number higher to 53%, which makes users even more likely to pick them, and so on. And then there are the aforementioned rogues, who intentionally pick an underdog they know nothing about simply because of the thrill of contradicting the mainstream opinion. Once the data on users’ choices is available to the users themselves, decisions can quickly change—in video games, bracket pools, and surely other fields as well.

Image credit: Kawakami, Mark. “Using YUI 3 to Build the Yahoo! Sports Tourney Pick’em Game.” 19 Mar. 2010. Web. Accessed 18 Feb. 2015. <http://yuiblog.com/blog/2010/03/19/tourney-pickem/>