Page - 70 - in Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources

3. ModelingMicrowaveHeating N1,1 N1,2 N2,1 N2,2 N2,3 N3,1 Input Layer Hidden Layer Output Layer W11,1 W21,1 YNN,1 N1,3 W13,4 Bias = 1 N3,2 YNN,2 W22,4 Bias = 1 YNN,2 uNN,1 uNN,2 Figure3.6. Diagramofrecurrentneuralnetwork are achieved by the network, and the corresponding process is called learningor training. As stated in [WF05], training is the mindless kind of learning. In this dissertation, they are assumed to be equivalent. A more formal defi- nitionof learning isgiven in [Hay98]: Learning is a process by which the free paramters of a neural network are adapted through a process of stimulation by the en- vironment inwhich thenetwork is embedded . Accordingtothedefinition,all freeparameters inneuralnetworkscan be modified during learning. There are mainly two types of learning. Oneissimilarwiththeprocessmentionedinthelastparagraph,where the topology of the network is fixed and only weights are adjusted. The second type of learning involves the modification of the topol- ogy of the neural network [SM04] [EUDC07]. It increases the num- ber of nodes in hidden layers and evolves weights of each link at the same time, according to predefined principles until the performance 70