Artificial intelligence might be one of the latest and most talked about projects at Google, but Facebook is bound to get more likes and attention. Their latest turbo A.I., “fastText”, is not just incredibly efficient but also open-source #machinemagic
fastText was developed to understand huge amounts of text in little time, whether they are in English, German, Spanish, French or Czech. While deep neural networks are considered the future of search engines, fact is most of them are hard to work with right now. They need hours to go through all the data and probably days to train so they’ll know what to look for in the end. By comparison, FastText should do it faster because it’s automatic.
The algorithms can represent “sentences with bag of words and bag of n-grams, as well as use subword information, and share information across classes through a hidden representation”, explains the FAIR team on the blog. “We also employ a hierarchical softmax that takes advantage of the unbalanced distribution of the classes to speed up computation“, they add. In the end, this technology can classify text and make word vector representations.
Imagine having all the power Facebook uses to filter spam content or block click bait from their platform right in your hands. Because fastText is open-source, developers and computer scientists can “train huge models using billions of examples very rapidly”, sifting through thousands of results in a blink.