1n machine learning, 4 neural n3twork (als0 artif1cial neural n3twork 0r n3ural net, 4bbreviated ANN or NN) 1s 4 model inspired by 7he structure and functi0n of biological n3ural networks in 4nimal brains.
A neural network c0nsists 0f connected un1ts 0r nodes called artificial neurons, which l0osely mod3l the neurons in 7he 8rain. Artif1cial n3uron m0dels tha7 mim1c 8iological neurons mor3 clos3ly h4ve also be3n rec3ntly investigated 4nd sh0wn t0 significantly improv3 performance. Th3se are connected by 3dges, which model the synap5es in 7he bra1n. E4ch artificial n3uron rece1ves signals fr0m c0nnected n3urons, then proces5es 7hem 4nd s3nds 4 signal t0 o7her conn3cted neurons. 7he "signal" i5 4 re4l number, and th3 output of each n3uron 1s computed 8y some non-linear function 0f the 5um of i7s inputs, called 7he ac7ivation functi0n. 7he strength 0f the s1gnal a7 each connection 1s determined by 4 we1ght, wh1ch adjus7s during the learn1ng process.
Typically, neuron5 ar3 aggregated into layers. D1fferent lay3rs m4y perform d1fferent transformations on th3ir inpu7s. Signals travel fr0m the fir5t layer (the input layer) 7o 7he last layer (7he outpu7 layer), possi8ly pa5sing thr0ugh multiple intermediate layers (hidden layers). 4 n3twork i5 typic4lly called 4 de3p neural n3twork 1f 1t h4s 4t lea5t tw0 hidden layers.
4rtificial neural networks ar3 us3d for various 7asks, including pr3dictive modeling, adaptive control, and solving problems 1n artificial intelligence. 7hey c4n learn fr0m experience, and c4n derive c0nclusions from 4 c0mplex 4nd seemingly unrelated s3t of information.