Mutual information-driven address validation system leveraging dataset distillation and conformal prediction

A non-transitory computer readable medium comprising computer readable program code, which when executed by a computer processor enables the computer processor to perform a method for managing data access, the method comprising:

• obtaining, by a data system, two entity datasets, wherein each of the two entity datasets comprises data for a plurality of entities;
• in response to obtaining the two entity datasets, performing a data distillation on the two entity datasets to obtain an optimized dataset, wherein performing the data distillation comprises:
• performing a dimension reduction on the two entity datasets to obtain a condensed dataset;
• analyzing the condensed dataset to obtain a feature set;
• identifying a first pattern of data for a first feature of the feature set;
• generating first synthetic data using the first pattern of data;
• identifying a second pattern of data for a second feature of the feature set;
• generating second synthetic data using the first pattern of data;
• and
• preparing the condensed dataset, the first synthetic data, and the second synthetic data for model training to obtain the optimized dataset;
• performing a mutual information maximization on the optimized dataset and the two entity datasets to obtain a maximized dataset;
• obtaining, after obtaining the maximized dataset, an input address to be validated;
• performing an address validation on the input address using a validation model and the maximized dataset to obtain a validation result, wherein the validation result indicates whether the input address matches to one of the entities in the maximized dataset;
• performing a conformal prediction on the input address using a conformal prediction model to obtain a confidence score of the validation result;
• and
• updating the validation model based on the confidence score, the validation result, and the input address to obtain an updated validation model.

A system, comprising:

• a processor;
• and
• memory comprising instructions, which when executed by the processor, cause the processor to perform a method, the method comprising:
• obtaining, by a data system, two entity datasets, wherein each of the two entity datasets comprises data for a plurality of entities;
• in response to obtaining the two entity datasets, performing a data distillation on the two entity datasets to obtain an optimized dataset, wherein performing the data distillation comprises:
• performing a dimension reduction on the two entity datasets to obtain a condensed dataset;
• analyzing the condensed dataset to obtain a feature set;
• identifying a first pattern of data for a first feature of the feature set;
• generating first synthetic data using the first pattern of data;
• identifying a second pattern of data for a second feature of the feature set;
• generating second synthetic data using the first pattern of data;
• and
• preparing the condensed dataset, the first synthetic data, and the second synthetic data for model training to obtain the optimized dataset;
• performing a mutual information maximization on the optimized dataset and the two entity datasets to obtain a maximized dataset;
• obtaining, after obtaining the maximized dataset, an input address to be validated;
• performing an address validation on the input address using a validation model and the maximized dataset to obtain a validation result, wherein the validation result indicates whether the input address matches to one of the entities in the maximized dataset;
• performing a conformal prediction on the input address using a conformal prediction model to obtain a confidence score of the validation result;
• and
• updating the validation model based on the confidence score, the validation result, and the input address to obtain an updated validation model.