In the startup world, companies seem to live and die by buzzwords. If your company isn’t using the latest “artificially-intelligent blockchain-based crypto-kazoo”, then you’re not startup sexy and investors’ attention will be lost. Machine learning is a collection of statistical methods that used to only be sexy in the data and computer science academic world. Over the last three years, Google searches for the term “deep learning” has grown by 770%. And more and more startups are being created with machine learning at their core. But many technology professionals may be wondering, “what does this all mean for me?”
Here are five problems that real developers and researchers are solving using machine learning (and similar statistical tools):
Netflix and YouTube use machine learning to recommend content to you that is similar to content that you liked in the past. If the system is good at it’s job, users will stay on the platform longer. The most successful recommendation engines are built using machine learning methods.
Financial companies are processing mountains of data every day. In order to detect fraudulent transactions, financial companies will employ teams of data analysts to detect anomalies. Analysts have found that various forms of machine learning can classify fraudulent transactions with remarkable accuracy. Researchers at companies like Cylance have been using machine learning in their anti-malware software to classify programs as “safe” or “malware”.
One way that IT is employing machine learning caught my attention. I found a guy on Hacker News that said his company was using deep learning to predict failing services on a server, and then auto-migrate the workload to another server before the first server crashes.
Sentiment analysis tries to guess the ‘mood’ of a given piece of text. At the startup I work for, I’ve used a sentiment analysis model to pull the sentiment out of online reviews on Google, Yelp, Facebook, etc. This allows our customers to get a really good understanding of their online reputation, and opens up more analytics than a simple 5-star scale. For example the model can tell them things like, “your customers enjoy the music you play,” and “your customers wish you were open later on Fridays,” just based off of their public online reviews.
Machine learning is most commonly used for classification. There are algorithms that can classify email as spam or not spam, algorithms that can classify images as trucks or cars, and algorithms that can classify songs into genres.
There are dozens of other interesting applications of machine learning (like self-driving cars), but most technology professionals won’t be working on incredibly advanced systems like that. Chances are, if you ever interact with machine learning, you’ll be working on a problem similar to one of the items above.
To learn more:
Intro to Machine Learning: https://www.udacity.com/course/intro-to-machine-learning–ud120
Google search trends for “deep learning”: https://www.google.com/trends/explore?date=all&q=deep%20learning
Malware detection using ML: http://www.darkreading.com/vulnerabilities—threats/researchers-enlist-machine-learning-in-malware-detection/d/d-id/1321423