US 12,229,664 B2
Optimization of artificial neural network (ANN) classification model and training data for appropriate model behavior
Prashanth Krishnapura Subbaraya, Bengaluru (IN); and Raghavendra Hosabettu, Bangalore (IN)
Assigned to Wipro Limited, Bangalore (IN)
Filed by Wipro Limited, Bangalore (IN)
Filed on Dec. 9, 2020, as Appl. No. 17/116,080.
Prior Publication US 2022/0121929 A1, Apr. 21, 2022
Int. Cl. G06N 3/08 (2023.01); G06F 18/23 (2023.01); G06F 18/24 (2023.01)
CPC G06N 3/08 (2013.01) [G06F 18/23 (2023.01); G06F 18/24 (2023.01)] 17 Claims
OG exemplary drawing
 
1. A method of optimizing artificial neural network (ANN) classification model and training data thereof for appropriate model behavior, the method comprising:
extracting, by a system for optimizing an ANN classification model and training data thereof, a plurality of entities and a set of domain specific entities from the training data for each of a set of classes of the ANN classification model, wherein the set of domain specific entities are representative of a set of domains;
determining, by the system, a plurality of model parameters of the ANN classification model based on the training data;
determining, by the system, missing data with respect to at least one of the training data and the plurality of model parameters based on the plurality of entities and the set of domain specific entities for each of the set of classes, wherein determining the missing data comprises:
clustering, by the system, the plurality of entities into a set of sequential clusters based on a domain associated with each of the entities for each of the set of classes;
grouping, by the system, one or more similar classes from the set of classes based on a degree of overlap among the set of sequential clusters;
determining, by the system, a superset of domain specific entities in the one or more similar classes;
determining, by the system, a cluster affinity value for each of the set of sequential clusters in each of the set of set of classes based on the set of domain specific entities and the superset of domain specific entities; and
determining, by the system, the missing data for each of the set of sequential clusters in each of the set of set of classes based on the cluster affinity value;
iteratively analysing, by the system, a relative advantage of a modified ANN classification model with a modified training data with respect to the ANN classification model with the training data, wherein the modified training data comprises one or more combinations of the training data and the missing data with respect to the training data, and wherein the modified ANN classification model comprises a plurality of modified model parameters generated by tweaking or removing one or more of combinations of the plurality of model parameters and the missing data with respect to the plurality of model parameters, and wherein the relative advantage provides for a criteria for determining a reliability of the ANN classification model and the training data with respect to the modified ANN classification model and the modified training data;
determining, by the system, an optimized ANN classification model and an optimized training data for appropriate model behavior based on the reliability; and
deploying, by the system, the optimized ANN classification model and the optimized training data in at least one of a computer vision task, an image recognition task, natural language processing task, speech recognition task, or a decision making task that represents the appropriate model behavior.