Ask HN: How does (or could) machine learning help you? | Hacker News Reader

Ask HN: How does (or could) machine learning help you?

15 pointstansey5 comments

Being a machine learning PhD student, I see a lot of really cool algorithms that are capable of amazing things, but typically they are only applied on toy problems. I'm curious in finding out if anyone has any experience with machine learning in the real world, either as a user, an implementor, or both. This covers both big data and more reinforcement learning approaches, offline and online methods, and personal and business life use cases. Alternatively, if you have a problem that you think machine learning can solve, but hasn't yet, let's hear it!*

* - Let's stay out of the realm of "find me things that I like" or "tell me news that I really want to hear". These tend to drift quickly into philosophical debates of "well you need some things you don't like too!" and such. I'm more interested in hearing more concrete problems. Maybe something you have to do by hand now, for instance.

An old friend of mine is using machine learning techniques to predict levels of drought occurrence ranging from a few dry weeks to a hypothetical dust bowl.

It sounds purely educational and research driven, but his goal is to study drought's influence on the commodities markets, so there's a potentially lucrative end goal.

http://www.datawrangling.com/how-flightcaster-squeezes-predi...

http://www.infoq.com/articles/flightcaster-clojure-rails

It's used for lots of stuff. Predictive typing, market segmentation, trend analysis and prediction, computer vision, medical tomography, process optimization...

See McKinsey's report:

http://www.mckinsey.com/mgi/publications/big_data/pdfs/MGI_b...

Suggested rule of thumb: the more a company uses the words "big data," the less likely they are to actually implement machine learning algorithms in their day-to-day.

"Big data" is, for the most part, large scale business analytics. You'll analyze Apache logs loaded onto a Hadoop cluster, clean the logs with a scripting language, use Hive/SQL to query your desired data, then analyze the result using R/Numpy.

Sorry, this has been an issue recently (exacerbated by Strata NY).

Reply on news.ycombinator.com