Best first search considering difference between scores
First Claim
Patent Images
1. A method of searching a lattice for problem-solving in artificial intelligence using a data processor, the lattice including nodes having present and next nodes each having node scores, and the lattice having depth and breadth, said method comprising the steps of:
- (a) inputting pattern data sets using an input device and storing the pattern data sets in a memory;
(b) determining candidate patterns for each of the input pattern data sets;
(c) generating pattern scores for the candidate patterns based on a similarity between the pattern data sets and the candidate patterns using the data processor to determine which candidate patterns are assigned to the nodes of the lattice so that the candidate patterns for each of the pattern data sets are arranged at a specific depth of the lattice in order of the breadth of the lattice according to the respective pattern scores thereof;
(d) calculating differences between the node scores of the nodes with the same depth using the data processor;
(e) calculating a search priority index for each of the nodes corresponding to the candidate patterns having a search priority based on the differences of the node scores of the nodes with the same depth using the data processor; and
(f) searching the lattice according to the search priority indexes and finding patterns corresponding to the pattern data sets using the data processor.
0 Assignments
0 Petitions
Accused Products
Abstract
A best first search for problem-solving in an artificial intelligence system employing a novel search priority index is disclosed. The search priority index is calculated based on a difference between scores of a node and the next node in breadth. Searching steps required to attain a solution can be reduced by employing the search priority index.
80 Citations
18 Claims
-
1. A method of searching a lattice for problem-solving in artificial intelligence using a data processor, the lattice including nodes having present and next nodes each having node scores, and the lattice having depth and breadth, said method comprising the steps of:
-
(a) inputting pattern data sets using an input device and storing the pattern data sets in a memory; (b) determining candidate patterns for each of the input pattern data sets; (c) generating pattern scores for the candidate patterns based on a similarity between the pattern data sets and the candidate patterns using the data processor to determine which candidate patterns are assigned to the nodes of the lattice so that the candidate patterns for each of the pattern data sets are arranged at a specific depth of the lattice in order of the breadth of the lattice according to the respective pattern scores thereof; (d) calculating differences between the node scores of the nodes with the same depth using the data processor; (e) calculating a search priority index for each of the nodes corresponding to the candidate patterns having a search priority based on the differences of the node scores of the nodes with the same depth using the data processor; and (f) searching the lattice according to the search priority indexes and finding patterns corresponding to the pattern data sets using the data processor. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17)
-
-
18. A method of symbol recognition using a data processor, comprising the steps of:
-
(a) inputting a pattern data set into the data processor with the pattern data set representing an input symbol to be recognized; (b) identifying with the data processor candidate patterns for the pattern data set; (c) generating by the data processor similarity scores for the candidate patterns; (d) assigning by the data processor the candidate patterns to nodes of a search tree with level depth and left to right position based on the similarity scores; (e) determining by the data processor score differences between similarity scores of nodes at a same search tree level; (f) calculating by the data processor search priority indexes according to the score differences; and (g) determining by the data processor a candidate pattern best matching the pattern data set to recognize the input symbol by searching the search tree using the search priority indexes.
-
Specification