Ex Parte SCHLANG et al - Page 2




            Appeal No. 1999-1856                                                                              
            Application No. 08/686,792                                                                        


                                               BACKGROUND                                                     
            The appellants’ invention relates to a learning process for a neural network.  An                 
            understanding of the invention can be derived from a reading of exemplary claims 15               
            and 16, which are reproduced below.                                                               
                   15.   A method for controlling an industrial process using a neural                        
                   network, the industrial process having at least one time-variable                          
                   parameter, the neural network including a control network and a                            
                   background network, the method comprising the steps of:                                    
                         training the control network using current process data to generate                  
                   a current process model;                                                                   
                         training the background network using representative process data                    
                   to generate an averaged process model of the industrial process over a                     
                   predetermined time period;                                                                 
                         determining a parameter of the at least one time-variable                            
                   parameter of the industrial process using the current process model, and                   
                         controlling the industrial process as a function of the determined                   
                   parameter.                                                                                 

                   16.   A method for controlling an industrial process using a neural                        
                   network, the industrial process having at least one time-variable                          
                   parameter, the neural network including a control network and a                            
                   background network, the method comprising the steps of:                                    
                         training the control network using current process data to generate                  
                   a current process model;                                                                   
                         training the background network using representative process data                    
                   to generate an averaged process model of the industrial process over a                     
                   predetermined time period;                                                                 


                                                      2                                                       





Page:  Previous  1  2  3  4  5  6  7  Next 

Last modified: November 3, 2007