NEURAL NETWORKS

Artificial Nurl Ntwrk (ANN) r bilgill inspired. Specifically, th brrw ideas frm th mnnr in whih the humn brin wrk. The humn brin i md f special ll lld nurn. Etimt f th numbr f nurn in a humn brin cover a wid rng (up t 150 billion), nd thr are mr thn a hundrd diffrnt kinds f nurn, separated int groups called networks. Eh ntwrk ntin vrl thousand neurons tht r highly intrnntd. Thu, th brin n be viwd as a lltin f neural networks.

Today's ANN, whose litin i rfrrd t nurl computing, u a vr limitd t f nt frm bilgil neural systems, th gl is t imult miv parallel processes tht invlv ring lmnt interconnected in network architecture. The artificial nurn riv inut nlgu t the ltrhmil impulses bilgil neurons riv frm thr nurn. The utut f th rtifiil nurn rrnd t ignl sent out from a bilgil neuron. Thi rtifiil signal can be hngd, like th ignl frm th humn brin. Nurn in n ANN riv infrmtin frm thr nurn r frm xtrnl ur, trnfrm r process th infrmtin, nd pass it n t other nurn or xtrnl outputs.

Th manner in which n ANN processes infrmtin dnd n it structure nd n th lgrithm ud t process the infrmtin.

The value f neural ntwrk technology inlud it ufuln fr pattern rgnitin, learning, and the interpretation f incomplete nd "noisy" inputs.

Nurl ntwrk hv th tntil to provide some f the humn characteristics of problem lving tht r diffiult to imult uing the lgil, nltil thniu of DSS r even xrt tm. One f th hrtriti i ttrn rgnitin. Neural ntwrk n nlz lrg untiti f dt to establish ttrn nd characteristics in itutin whr th logic nd rules are nt knwn. An xml wuld b ln litin. B rviwing mn hitril of lint' questionnaires and th "yes or n" decisions md, th ANN n rt "ttrn" r "profiles" f applications tht huld be rvd r dnid. A nw application can thn matched b th mutr against th pattern. If it m l nugh, the computer lifi it as a "yes" or "n"; otherwise it g to a human fr a diin. Nurl ntwrk r especially useful for financial applications such as dtrmining when to bu or ll tk, rditing bnkrut, nd rditing xhng rt.

Beyond its role an ltrntiv muting mechanism, nd in data mining, neural muting can be combined with thr mutr-bd infrmtin tm t rdu wrful hbrid tm.

Nurl muting i emerging as n fftiv technology in pattern rgnitin. Thi capability i being trnltd t mn litin nd is sometimes intgrtd with fuzz logic.

Dettagli