Applying Artificial Neural Networks
An Artificial Neural Network (ANN) is a machine-learning algorithm for function approximation. Essentially an ANN consists of a bunch of input nodes with edges to a bunch of output nodes. There can also be hidden layers between the input and output nodes like this: http://goo.gl/zk67Ir. Unlike most of the networks we have discussed so far in class, the edges in an ANN have a direction. Each node adds it’s inputs (the edges pointing to the node), calculates a value based on the inputs, and outputs the value to the next layer of nodes (the edges pointing away from the node). Each of the edges has a certain weight attached to it, which is essentially how much influence the output of one node has on the input of the other node. The ANN works by adjusting the weights of edges to produce better results based on the data you “teach” the network with.
ANNs are very good at dealing with complex situations that have many inputs. One such case is diagnosing diseases. Often doctors are flooded with information and data from the patient and it is difficult to distinguish important from non-important information. According to the IBM website, many misdiagnoses are due to doctors not having the correct information. Papers have suggested the use of ANNs in to help diagnose disease since the 1990s. However, more recently IBM has created a product called WatsonPaths, which uses machine learning algorithms such as ANNs to help doctors. WatsonPaths both provides information it deems relevant to the doctor and draws its own conclusions. It is able to analyze far more information than a person is and therefore is able to potentially create better diagnoses, or at least help guide doctors in promising directions. Through applications like WatsonPaths, networks just like the ones we are discussing in class are saving lives around the world.
Source:
Additional Information:
http://en.wikipedia.org/wiki/Artificial_neural_network
http://www.sciencedirect.com/science/article/pii/S0140673695918043#