초록 ▼

The present invention discloses a method for building a self-learning evidential reasoning system from examples. In this invention a hierarchical model structure for the self-learning evidential reasoning system is defined. After the model structure has been defined examples are supplied by experts.

The present invention discloses a method for building a self-learning evidential reasoning system from examples. In this invention a hierarchical model structure for the self-learning evidential reasoning system is defined. After the model structure has been defined examples are supplied by experts. The examples are entered directly into example spreadsheets and then used to train the model structure. The model structure is then trained to minimize error between model output and the desired output. The error is minimized by using a gradient descent optimization. The model structure is then put into production and then used to make a recommendation. The self-learning evidential reasoning system can be used in various fields, such as analyzing the risk of financial service applications.

대표청구항 ▼

[ I claim:] [1.] A method for building a self-learning evidential reasoning system comprising the steps of:defining a hierarchical model structure for the self-learning evidential reasoning system, the hierarchical model structure having a plurality of processing nodes each having a set of inputs an

[ I claim:] [1.] A method for building a self-learning evidential reasoning system comprising the steps of:defining a hierarchical model structure for the self-learning evidential reasoning system, the hierarchical model structure having a plurality of processing nodes each having a set of inputs and an output, wherein outputs from lower level processing nodes are propagated as inputs to higher level processing nodes until a final output is generated from the highest level processing node in the hierarchical model structure;specifying a number of attributes for each of the plurality of processing nodes, each attribute describing a factor to be considered by that processing node;specifying a number of possible linguistic evidential values for each of the attributes;collecting a plurality of examples for each of the processing nodes in the hierarchical model structure, each of the plurality of examples having a set of inputs and a corresponding desired output; andtraining the processing nodes with the plurality of examples such that the model structure learns to generate a final output from a set of inputs, the model structure being trained to have minimal error between the final output and the desired output for the plurality of examples, the training comprising the steps of:providing a linguistic evidential value for each of the inputs in the plurality of examples;translating the linguistic evidential values into an evidential numeric value;determining a weighting function for each processing node for aggregating the evidential numeric values into an evidential numeric aggregate value; andoptimizing the weighting functions for each of the processing nodes in the model structure until there is minimal error between the final output and the desired output for the plurality of examples.