Machine Learning

Machine learning trains computers to learn from experience.

Machine learning trains computers to learn from experience.

Computers are great at following very specific directions. If an email has the words “Nigerian oil minister” and offers you “$3,000,000” it’s probably spam. But what about a Kuwaiti oil minister, or ten million dollars? Machine learning is the process that allows computers to recognize patterns and improve from experience.

Google can recognize spam thanks to a machine learning process called supervised learning. First, it’s fed millions and millions of emails that it knows are spam. Each time you mark something as spam, you aren’t just getting it out of your inbox – you’re also alerting Google to take a hard look at that email and compare it to the rest of the spam they know about. This training data allows Google’s algorithms to learn words and phrases associated with spammy emails.

Even if spammers change their tactics or email you new scams, it doesn’t take long for Google to recognize them as something you don’t want to read. Machine learning is useful for systems that change over time or processes need to adapt to new situations.

Driverless cars, chess-playing computers, and automatically tagging friends in photos are all examples of machine learning at work.

Cocktail Party Fact

In 2006, Netflix held a $1,000,000 competition to improve its recommendation system. An algorithm would examine your past movie ratings and try to predict your ratings for movies you haven’t seen yet. It’d recommend the ones it thought you’d rank the highest.

Eventually a team improved Netflix’s algorithm by 10% and walked away with the prize. Netflix never ended up using the new algorithm, though! The new one was too complex and not cost-effective for their system, so they stuck with the old one.

Machine learning is a tough field that takes a lot of data and computing power, and when it comes down to it sometimes just enough is more.