Data Mining and Knowledge Discovery, Neural Networks in

摘要: Activation function The activation or transfer transforms the weighted inputs of a neuron into an output signal. functions often have “squashing” effect. Common used in neural networks are threshold, linear, sigmoid, hyperbolic, and Gaussian. Artificial network An artificial is system composed many simple, but highly interconnected processing nodes (named neurons) which operate parallel collectively. It resembles biological nervous systems two basic functions: (1) Experiential knowledge acquired through learning process can be retrieved again later. (2) stored strength (weights) connections between neurons. receives number inputs, may either external to outputs other Each input connection assigned weight, similar synaptic efficacy neuron. sum compared against level (threshold) determine value Feedback In feedback recurrent networks, signals flow both directions. dynamic such that they state changing continuously until it reaches equilibrium point. Learning rule describes way trained, i.e., how its free parameters undergo changes fit training data. Feedforward organized one more layers units. feedforward network, signal allowed only, from outputs. There no loops, layer do not affect inputs.