Neurons are complex, but even a highly simplified model of a neuron when connected with others in an appropriate network, can perform powerful computations. A biological neuron receives information from as many as thousands of other neurons through synaptic connections and passes on signals to as thousands of other neurons. The synapse, or connection between neurons, mediates the “strength” with which a signal crosses from one neuron to another. Artificial “neural” circuits have been built from simple electronic components: operational amplifiers replace the neurons, and wires, resistors and capacitors replace the synaptic connections. The output voltage of the amplifier represents the activity of the model neuron, and currents through the wires and resistors represent the flow of information in the network.
Both the simplified biological model and the artificial network share a common mathematical formulation as a dynamic system - a system of many interacting parts whose state evolves continuously with time. The manner in which a dynamic system evolves depends on the form of the interactions. In any neural network the interactions result from the effects one “neuron” has on another by virtue of the connection between them. Thus it is not surprising that the behavior of the neural circuits depends critically on the details of the connections, and the strengths of each.
The computational behavior exhibited by neural networks is a collective property that results from having many computing elements act on one another in a richly interconnected system. The collective properties can be studied using simplified model neurons based on operational amplifiers, resistors, and capacitors.
Further reading:
Neural Network
Silicon neurons
Biological neural network
No comments:
Post a Comment