We will begin by providing a simple description of a biological neuron, and thus motivate a model called an artificial neuron.

1.1 Biological neurons

For pedagogical purposes, we will separate the workings of a biological neuron into its input and its output.

1.1.1 Input

Artificial Neural Networks (ANNs) are based upon our limited understanding of biological neural networks. A biological neuron has many inputs and one output. The signals which travel along the inputs and outputs are electrochemical and consist of trains of pulses. The frequency of the pulses seems to be the important factor, with a low frequency corresponding to a weak signal, and a high frequency corresponding to a strong signal. The output of the neuron depends on the strengths or values of all of the input signals. However, it is more sensitive to some of the inputs than to others. Thus each input has a connection strength which determines how much influence that input has on the neuron. Furthermore, some of the inputs have an inhibitory effect, which means that instead of increasing the output of the neuron, they reduce it. The neuron reacts to the net effect of all inputs.

1.1.2 Output

The output of the neuron is a function of its net input, and the output rises as the input rises. However, since the output is a pulse train, where frequency equals signal strength, it cannot fall below zero. There is also an upper limit to how fast the neuron can generate pulses, hence the neurons's output signal has an upper limit. Lastly the input region over which the output goes from its minimum value to its maximum value is not fixed, and may occur either around a small net input or a large net input.

The relationship between the net input and neuron output is approximately the same as that shown below.

1.1.3 Summary

The above is a sketch of two biological neurons. The one on the right has several of its inputs coming from the one on the left. If these neurons were in a brain then the left neuron's output would go to many other neurons, and the right neuron would receive inputs from many other neurons.

Back to top. >>>

Continue with the tutorial. >>>