The concept of a neural network (NN) is to develop a
simple mathematical model of the human brain. A neural network is expressed
as the network model that consists of neurons and weights which connect each
neuron. This network is the function that derives the result corresponding
to the teaching process from input data.
The learnig process adjusts the weights of connections
between neurons for the purpose of getting the output relevant to specified
input signals. If you want to make the property estimation system, the molecular
geometry information are used for the input data. The training data, such
as well-known molecular geometry information and properties, are used for
the learnig process.
The trained neural network is the function that extracts
the relationship between molecular geometry information and physical property
values. You can calculate the unkown molecular properties which are not
used for the learnig process and get it's estimation properties based on
the relationship that is learned.
Because the neural network system is represented by
a non-linear function, it can provide efficient estimation functions even
if there are difficult properties, which can't be represented by a linear
function like multiple regression.
HOME
Previous Page
TOP