Evolutionary Optimization of Data Transformation Models
AssigneeMastercard International Incorporated
InventorsSiddharth Vimal, Gaurav Dhama, Kanishka Kayathwal, Nishant Kumar
Life of this patent
Estimating the term…
Abstract
Various implementations relate to evolution-based optimization of data transformation models. A server system applies a plurality of encoders to input data to generate transformed input data instances. The transformed input data instances are applied to a target machine learning model to generate corresponding decision scores. For each encoder, a fitness score is computed based on a corresponding decision score. A subset of the encoders is selected based on the fitness scores, and evolved encoders are generated by modifying parameters associated with the selected subset.
Claims20
A computer-implemented method, comprising:
- applying, by a server system, a plurality of encoders to input data to generate corresponding transformed input data instances;
- applying, by the server system, the transformed input data instances to a target machine learning model to generate corresponding decision scores;
- computing, by the server system, for each of the plurality of encoders, a fitness score based on the corresponding decision scores generated by the target machine learning model for a corresponding transformed input data instance;
- selecting, by the server system, a subset of the plurality of encoders based on the fitness scores;
- and
- generating, by the server system, a plurality of evolved encoders by modifying parameters associated with the subset of the plurality of encoders.
A server system, comprising:
- a communication interface;
- a memory comprising executable instructions;
- and
- a processor communicably coupled to the communication interface and the memory, the processor configured to execute the instructions to cause the server system, at least in part, to:
- apply, by a server system, a plurality of encoders to input data to generate corresponding transformed input data instances;
- apply, by the server system, the transformed input data instances to a target machine learning model to generate corresponding decision scores;
- compute, by the server system, for each of the plurality of encoders, a fitness score based on the corresponding decision scores generated by the target machine learning model for a corresponding transformed input data instance;
- select, by the server system, a subset of the plurality of encoders based on the fitness scores;
- and
- generate, by the server system, a plurality of evolved encoders by modifying parameters associated with the subset of the plurality of encoders.
A non-transitory computer-readable storage medium comprising computer-executable instructions that, when executed by at least a processor of a server system, cause the server system to perform a method comprising:
- applying a plurality of encoders to input data to generate corresponding transformed input data instances;
- applying the transformed input data instances to a target machine learning model to generate corresponding decision scores;
- computing for each of the plurality of encoders, a fitness score based on the corresponding decision scores generated by the target machine learning model for a corresponding transformed input data instance;
- selecting a subset of the plurality of encoders based on the fitness scores;
- and
- generating a plurality of evolved encoders by modifying parameters associated with the subset of the plurality of encoders.
Dependencies are inferred from the claim text.