Applied Math & ML #1: Neurons
Introduction to artificial neural networks in machine learning…starting with neurons!
Intro: What Deep Learning is Modeled After
Through trial and error, evolution has created what we call “life”. A life form is a system of complex operations that work in harmony. One of the most advanced forms of life is the human being.
Rather than just physical abilities, our level complexity mainly resides in our brain. Our brain is able to work in harmony with the rest of our body thanks to the nervous system. The nervous system allows for the brain and body to communicate. The communication is done through signals (which can be either electrical or chemical) between specific cells called neurons.
A neuron accepts messages from other cells through its dendrites. Those messages from other cells are inputted into the nucleus of the neuron. The neuron then tries to evaluate what’s actually going in the world around it.
Just like when several close friends give you advice on what you should do, you weigh the advice of certain friends higher than others.
You may have two friends, one is overly dramatic and the other is extremely passive. Because their polar opposite of each other, you would weigh them evenly and make a decision based on the middle ground.
Now, if one is overdramatic and the other is normally rational, you would give the overdramatic friend’s input less weighting (near zero percent) and the normally correct friend more weighting (near 100%).
The neuron is very similar — except each neuron has thousands of close friends. It gets messages (called “inputs”) and it has previously set weighting (how much they consider the accuracy of the message) for each message source. After analyzing all the inputs against their weightings, the neuron then makes a decision: either to send or not send out its own message (sent through the axon), and how many times/how “loud” should it scream this message (how dramatic/serious to make it).
When you want to move your arm, it starts with one neuron “screaming” then it exponentially becomes a billion neurons screaming. These neurons scream from your brain and down your spine until it reaches the neurons in your arm. Then the neurons in your arm scream at the muscle cells in order to induce movement. ← (Extremely simplified version of the process human motor actions) This overall system of messages from the neurons in your brain to the rest of the body is called the nervous system.
In the process of advancing technology, we often try not to reinvent the wheel. So we looked inward when trying to make robots as smart as ourselves, and the nervous system was our muse (or at least our best muse so far).
The software version of a neuron is a perceptron. In the most basic implementation, it’s a function that takes two multidimensional arrays (one for inputs and one for weightings), does some mathematical calculations (which we’ll later explore), and spits out an answer.
However, a perceptron by itself isn’t that powerful. Similar neurons in the brain, it’s the combination of a large number of perceptrons that allows for greater functionality and accuracy. This architecture of numerous perceptrons is called a multi-layer perceptron (MLP).
To be continued in Applied Machine Learning #2…
Thanks for reading! For more articles covering coding, music, dating, and the overall urban nerd lifestyle follow The Urban Nerd publication or follow me on Snapchat (https://snapchat.com/add/the-urbannerd) or Twitter (https://twitter.com/TheUrbanNerd). Preferably SC!
