Usually computer code functions in a linear way, they are driven by simple 0s and 1s, true and false statements. Easy, accurate in its own statements and the way it has always been.
But what if we need our computer, smartphone or any other system to interpret connotation or data that is not as easy interpretable by a computer functioning in true or false statements?
What if we’d like the computer to interpret data more like our brain does?
Neural networks, as you might suspect by the terminology is a solution for our computer to think more like a brain.
Our brain and how it interprets stuff
Neurons in our brain process information so we understand and interpret what we perceive. The extremely large interconnected network of neurons collect information.
This is a neuron in our brain:
A neuron does not work in a binary code, it collects inputs from other neurons, sums all the inputs and if the resulting value is greater than a certain value, it fires a signal.
So instead of a certain number of combinations of 0s and 1s, there are x number of possibilities:
Neural networks: a computer that works as a brain
Neural networks function just like our brain. X number of possibilities are summed and based on this number, an action is determined.
This implies:
- your computer has more computing power
- more options to investigate and return the right answer and
- you’ll obtain multiple forms of input to obtain a correct output
Let’s make it a little bit more concrete. Imagine a sentence with the word ‘heart’ in it.
Given the context of the word, in this case the rest of the sentence, a computing system in a neural network will be able to interpret if we’re referring to a human physical heart, an animal physical heart, heart as a reference to love or any other meaning the word ‘heart’ could have.
This opens up a range of new possibilities. Computers are able to interpret data in a complete new way.
But… using simple neural network algorithms they’re still only up to 97% accurate while doing this and we demand a higher accuracy from our computer’s results.
Multi layer neural networks: train the brain
Multi layer neural networks show a 99% or more accuracy in data interpretation.
These multi layer neural networks have to be trained: all possible perceptions have to be known.
Data sets of raw data are matched with what should be interpreted.
This means you need servers to generate data for your neural network.
The system must of course learn somehow what data should be matched. This training can be done manually which will result in an algorithm.
For example… you’ve guessed it: Facebook algorithms. Images of people tagged on Facebook are recognized thanks to multi layer neural networks. And we’re all manually helping Facebook’s algorithm. Nice job!
Our future is personalized
What’s in it for us, you ask?
Personalization. Of course.
Android 8 and the new Google phone with its chip dedicated to neural networks and library or backup CPU will personalize our news feeds, images on our smartphone, ads,…
So not just online content will be personalized, even our local files will be accurately designed to our preferences.
Wanna know more?
Watch this great introduction to neural networks:
source: How Deep Neural Networks Work - Brandon Rohrer
