| Introduction | | | | input information to the output. |
| Human beings and machines may have much in | | | | 3. One output layer to give the answers, in this |
| common. Humans invent new machines all the | | | | case yes/no if the music notes correspond to a |
| time and they say that the first machine ever | | | | certain song. |
| invented was the wheel. | | | | How does the ANN learn? |
| From far dark ages in history to nowadays | | | | The ANN learns by iterations or repetitions, and |
| computers age and at the core of every machine | | | | this iterations are called epochs. |
| there is One and Zero.Known as the binary | | | | So for each learning epoch in the ANN there is: |
| system, in fact, when you see the letter "A" in | | | | |
| your computer there are Ones and Zeroes behind | | | | 1. Feed input data |
| it. When you use the most complex software or | | | | 2. Propagate signal through layers |
| surf to your favourite web site, there are Ones | | | | 3. Give an output |
| and Zeroes there too. Spiritual people say that | | | | Well then, if we don't tell the net when to stop |
| the Universe is made of nothing (zero) and | | | | the loop can go on forever. This flow needs to be |
| something (one), that the Universe is made | | | | more elaborated by setting stopping conditions |
| mostly of emptiness. | | | | somewhere, sometime when it is for sure that |
| The film A.I. has created a myth in people's mind | | | | the net has learned. |
| perceiving artificial intelligence as some kind of | | | | Like in the biological model the neurons transmit |
| "magic" of the technology. Also, in the old sci-fi | | | | the electrical impulses through layers of neurons in |
| films we always see machines, those gigantic | | | | the brain till there is a desired output. |
| computers who develop independent free will to | | | | The most known ANN model is called multilayer |
| take control over humans. Not so nice picture, | | | | backpropagation or multilayer perceptron, and a |
| huh? | | | | perceptron is simply a neuron that learns. |
| In this paper I will try to demystify the idea of | | | | Let's expand the learning model a little bit more |
| artificial intelligence by giving simple explanations | | | | by creating a stopping condition called the |
| and with no mathematics if possible, putting in | | | | minimum desired error (the ANN learns from its |
| your hands the simple truth: all there is behind is | | | | errors just like us! Well, ahem! sometimes..),so: |
| One and Zero. | | | | |
| Back in 1943 McCulloch and Pitts developed | | | | 1. Feed input data. |
| models of artificial neural networks (from now on | | | | 2. Propagate signal through layers from the output |
| ANN) based on their understanding of neurology, | | | | last layer backwards to the first hidden layer. This |
| those discoveries found out how neurons learn in | | | | is backpropagation. |
| the human brain: by transmitting electric impulses | | | | 3. Calculate current error. |
| through the synapses (connections) between | | | | 4. Ask: Is current error smaller than the minimum |
| neurons. | | | | desired error? Then give output and EXIT. |
| We could say that neurons in our brain are united | | | | 5. If current error is bigger: Go back to 1. |
| through a gigantic number of connections the | | | | This model is yet a very simple model as one |
| makes the whole act like an enormous almost | | | | could ask: what if the current error is never |
| infinite network. | | | | smaller than the minimum desired error? Then we |
| Well, this idea was transported to software | | | | can create a second stopping condition, the |
| research to create an algorithm, or method, that | | | | maximum number of iterations (epochs) allowed. |
| can learn like the brain does: through connections | | | | In step two (backpropagation) some necessary |
| and signal propagation through neurons. | | | | mathematics calculations are done to find out the |
| Our brain needs the input data, like reading, | | | | current error. |
| smelling, or hearing music, then the brain filters all | | | | This calculations are based upon the connections |
| through electrical impulses and waves. | | | | between layers. I am not going to deep in the |
| When one listens to only a few tunes he/she can | | | | formulas details I'm just going to give the idea |
| recognize the melody and tell the songs name | | | | behind it: |
| before the end of the play. | | | | My Actual Layer Data=My Previous Layer's |
| Here the input are the music notes and the | | | | Calculations. |
| output the song's name just recognized. Easy.. | | | | And the word Previous is very important here |
| In the same manner we can design an ANN: | | | | because it is the way that layers are connected |
| | | | to each other. |
| 1. Input | | | | Conclusion.. and what's really inside a neuron? |
| 2. Processing | | | | So far we've talked about neurons, networks, |
| 3. Output | | | | layers, input and output data, backpropagation and |
| But a single note will not be enough to recognize a | | | | epochs. |
| whole melody and so the ANN needs more input | | | | All these words are the usual terminology in all |
| data to learn before being able to give a valid | | | | ANN papers but this paper is different and I want |
| output. | | | | to talk about what is inside a neuron. |
| Why does the ANN need layers? | | | | Inside a neuron there is One or Zero and the |
| The web connections in an ANN are organized in | | | | output solution once the network has learned is |
| layers, and a layer contains from one to many | | | | given as One (true) or Zero (false). Of course |
| neurons, so, for the music problem the layer's | | | | there are ANNs that work with real numbers like |
| distribution is: | | | | 1.5672 but in most cases the input data is scaled |
| | | | close to Zero or One values to make sure that |
| 1. One Input layer containing information for the | | | | the best performances are given. |
| ANN to learn, let's say the music notes where | | | | After these very simple explanations Artificial |
| each note is a neuron. | | | | Intelligence is in your hands now and you can walk |
| 2. One to several hidden layers that will connect | | | | your way on. |