A recent article in the MIT Technology Review has caused some people to look at existing technologies in a new way. Search engines have been around for some time now. In fact, they date back to around 1982. And what’s really amazing is that they haven’t really changed much in all that time. As the article points out, they’re mainly focused on taking in text, running it through a database, and outputting results based on keyword matching. The biggest innovations there have come from having actual humans double check the results in order to help machines match questions and answers together. There’s little real intelligence involved, just simple pattern matching.
This is all set to change thanks to a field of artificial intelligence known as deep learning. This is being combined with another technique known as visual search in order to create the next generation of search engine. There’s two things which really sets this new type of search technology apart from existing techniques. The first is that it’s based on visual data such as images or pictures taken with one’s phone. And the second is that it constantly makes use of actual intelligent analysis. Even if that intelligence is of the artificial variety.
Obviously this is a lot more computationally taxing than the current style of simply sending text into a database. But various companies are experimenting with different solutions to the problem of how processor intensive the techniques are. A company called Slyce has one of the most promising takes on it though. They use a method of cloud computing called distributed processing in order to work with deep learning and visual search techniques.
Slyce then created a special API which allows programs to tie into their servers through the Internet. This essentially allows Slyce’s servers to act like a dual core processor. Less powerful devices like smartphones or online catalog systems can handle the taking of and trading off of images. But the actual processing and analysis of that visual data is handled by the virtual processors at Slyce. The end result is systems which can handle visual processing that might well be dozens or even hundreds of times more computationally intense than they would otherwise be able to process. And thanks to the nature of Slyce’s easy to use infrastructure it can be added to almost any existing program.