Machine Learning Challenges in 2018
Jun 14 2018

What Is Machine-Learning?

Machine-learning is a type of artificial intelligence that allows computers to “learn” from real-time and historical data. The surge-pricing models used by rideshare companies are a good real-world example of machine learning.
These models use machine-learning to figure out what price and conditions the market will bear so the company can optimize revenue and make their driver network more effective. Traditional software is only as good as the rules it is programmed with. It can’t do the pattern-recognition required to discern that in Chicago, people are very price sensitive, but they’ll wait for a car for more than 10 minutes, whereas in San Francisco, people will pay 20% more, but they’ll cancel if they see a wait-time of longer than 5 minutes. Static software can’t adapt to multivariable, changing, and, especially, unexpected conditions. Machine-learning models can.

The Challenges

For good data scientists, machine-learning models aren’t that hard to build. (And there are data science companies today trying to fill that gap already for enterprises.) What’s hard about machine-learning models is that they are very difficult—and expensive and time-consuming—to maintain. Add to the debugging, updating and rewriting you have to do with static software the unpredictable, sometimes hidden, and infinitely more complex challenges of models that are being influenced by vast streams of live data… This is the flip-side to machine learning’s great promise, and it’s where Datatron comes in.
Filed under Datatron Platform | Tagged: Machine Learning Challenges in 2018