Neural networks are modelled on the way the brain is believed to process data. It consists of a net of nodes joined by links. Without getting too deep in technical jargon, the Neural net has similarities to a good old fashioned Ontology where key terms are linked by a relationship to make a huge net of nodes all relating to each other in some way. In this instance a search engine finding the term “Desert” would have no confusion because it was joined by references to sand and dunes, therefore it could rule out Tiramisu etc. Naturally nets are very much more complex but surprisingly similar. The brain, like the computer, ignores slight spelling errors to a considerable degree regardless how much the owner of that brain may complain about spelling and grammar imperfections.
Rather than words, a modern Neural net could be linking any kind of concept, thing or capability.
For example in recognising a face of a person approaching, it might first recognise a physical thing, then notice the height width and movement as likely to be human, then look for hands and feet and make a classification. Moving on from that initial classification is a continual progression until it recognises John Schmidt from Bonn.
The interesting points are that it really does ape humans quite a bit. Just like us it quickly decides, this is human and begins searching for the hands. If the man has been in the wars and has no hands, it could however, bomb out spectacularly.
If you are a programmer you will recognise the performance benefits of knowing which programmes to start up in anticipation of being needed and which ones can be shut down for now. The downside is the potential for cognitive dissonance.
The efficiency of Neural nets as per Hinton and Kurtzweil is the efficiency of reusing knowledge instead of doing it again. E.G once a net understands what a nose is, it only has to compare the one it has to the new one on a cat and remember the difference, not the whole thing. On a good day, we humans do this, but specialisation, dogma and politics has robbed us of much of our thinking power when in formal situations. I always feel that this is why a long enjoyable walk often solves more problems than a working lunch.
The “Explained series” is planned to build into a trustworthy collection of explanations and commentaries that can be trusted to tell the story straight without any bias and attempt to make the subjects accessible to the layman. The latter is not always easy as some of these terms refer to genuinely complex subject matter, while others are simply too vague to pin down (there’s another word for that). There is also a limit to how far I can go in explaining every term when there are a lot of them, so I have to sometimes rely on your initiative to right click the offending word and look it up.
If you want an answer on something and you can’t find it easily, please use the comments section to just ask and I will appreciate not having to research the next topic.