Mini:Neural networks
The artificial neural network (NN) is a kind of computer software program that is loosely modeled after the more parallel neuronal structure of the brain, only more simple and on a much smaller scale.
Like the brain, the individual elements or units that comprise the NN are often called 'neurons'. Although these artificial neurons exist only as mathematical processing subroutines, they too are interconnected with synapses or 'weights', and contain thresholds that determine when they will “fire” or pass information to one another.
Each neuron may have several inputs, either from the data itself, or from other neurons. These inputs are multiplied by an associated weight. This weight corresponds to the strength of the connection to other neurons, and is analogous to the resting potential of the biological neuron, in that it helps determine the output or “firing” of the neuron. Then we take the sum of these weighted inputs, and apply a function to this sum. If the sum exceeds a certain threshold, the neuron fires off the signal to serve as inputs to the connected neurons. The simplified equation for the neuron above is therefore:
To create an artificial neural network, we must group these neurons into 3 types of "layers", with one “input” layer, one “output” layer, and at least one “hidden” layer.
The layers of neurons are connected in such a way that each neuron in one layer is able to “communicate” or send a signal to every neuron in the next layer. The first layer generally receives the input, directly from your data.
In a weather prediction neural network, the input layer may have variables such as the humidity, the barometric pressure, the high/low temperature range, and the cloud patterns. The middle or hidden layer contains neurons that will process data from the input layer, and internally store an abstract mapping of the data patterns. This layer allows the network an additional level of sophistication. The hidden layers process the data and pass it along to the output layer, which contains neurons to represent different possible output values. In our weather example, possible outcome may be simply “rain”, or “no rain”.
The weights carry input signals forward through the NN. Like the biological neurons in our brain, the real processing power of an NN is in the overall pattern of these weighted connections, not in any single neuron. These NNs can simulate the processing of a large quantity of data in parallel, rather than one instruction at a time. The NN is therefore a computational model, used to solve classification and pattern recognition problems.
 | This is a mini page for a popular search term. |
| Help to improve this mini by editing it! |