System and method for order-preserving encryption for numeric data
View Patent ↗A system, method, and computer program product to automatically eliminate the distribution information available for reconstruction from a disguised dataset. The invention flattens input numerical values into a substantially uniformly distributed dataset, then maps the uniformly distributed dataset into equivalent data in a target distribution. The invention allows the incremental encryption of new values in an encrypted database while leaving existing encrypted values unchanged. The flattening comprises (1) partitioning, (2) mapping, and (3) saving auxiliary information about the data processing, which is encrypted and not updated. The partitioning is MDL based, and includes a growth phase for dividing a space into fine partitions and a prune phase for merging some partitions together.
1. A computer-implemented method for preventing data reconstruction, comprising:
modeling input and target data distributions;
flattening original numerical values into a flattened dataset having substantially uniformly distributed values; and
transforming the flattened dataset into an encrypted dataset having values distributed according to the target data distribution;
wherein the modeling comprises:
partitioning data into buckets; and
modeling the data distribution within each bucket as a linear spline;
wherein the partitioning includes:
a growth phase wherein the data space is recursively split into finer partitions; and
a pruning phase wherein some buckets are merged into bigger buckets.
2. The method of claim 1 wherein new values are incrementally encrypted while leaving existing encrypted values unchanged.
3. A computer-implemented method for preventing data reconstruction, comprising:
modeling input and target data distributions;
flattening original numerical values into a flattened dataset having substantially uniformly distributed values; and
transforming the flattened dataset into an encrypted dataset having values distributed according to the target data distribution;
wherein the modeling comprises:
partitioning data into buckets; and
modeling the data distribution within each bucket as a linear spline;
wherein the partitioning further comprises creating two special buckets, one at either end of the domain of the input distribution, for encrypting values outside a current data range.
4. The method of claim 1 wherein the flattening comprises:
defining a mapping function for each bucket including a quadratic coefficient and a scaling factor; and
retaining all bucket boundaries, quadratic coefficients, and scale factors in a data structure.
5. The method of claim 1 wherein the transforming comprises:
defining a mapping function for each bucket including a quadratic coefficient and a scaling factor; and
retaining all bucket boundaries, quadratic coefficients, and scale factors in a data structure.
6. The method of claim 3 wherein a given plaintext value is mapped to a plurality of encrypted values to prevent an adversary from using duplicates to guess the distribution of a domain.
7. A general purpose computer system programmed with instructions for preventing data reconstruction, the instructions comprising:
modeling input and target data distributions;
flattening original numerical values into a flattened dataset having substantially uniformly distributed values; and
transforming the flattened dataset into an encrypted dataset having values distributed according to the target data distribution;
wherein the modeling comprises:
partitioning data into buckets; and
modeling the data distribution within each bucket as a linear spline;
wherein the partitioning includes:
a growth phase wherein the data space is recursively split into finer partitions; and
a pruning phase wherein some buckets are merged into bigger buckets.
8. The system of claim 7 wherein a given plaintext value is mapped to a plurality of encrypted values to prevent an adversary from using duplicates to guess the distribution of a domain.
9. A general purpose computer system programmed with instructions for preventing data reconstruction, the instructions comprising:
modeling input and target data distributions;
flattening original numerical values into a flattened dataset having substantially uniformly distributed values; and
transforming the flattened dataset into an encrypted dataset having values distributed according to the target data distribution;
wherein the modeling comprises:
partitioning data into buckets; and
modeling the data distribution within each bucket as a linear spline;
wherein the partitioning further comprises creating two special buckets, one at either end of the domain of the input distribution, for encrypting values outside a current data range.
10. The system of claim 9 wherein new values are incrementally encrypted while leaving existing encrypted values unchanged.
11. The system of claim 9 wherein the flattening comprises:
defining a mapping function for each bucket including a quadratic coefficient and a scaling factor; and
retaining all bucket boundaries, quadratic coefficients, and scale factors in a data structure.
12. The system of claim 9 wherein the transforming comprises:
defining a mapping function for each bucket including a quadratic coefficient and a scaling factor; and
retaining all bucket boundaries, quadratic coefficients, and scale factors in a data structure.
13. A data management service for preventing data reconstruction, comprising:
modeling input and target data distributions;
flattening original numerical values into a flattened dataset having substantially uniformly distributed values; and
transforming the flattened dataset into an encrypted dataset having values distributed according to the target data distribution
wherein the modeling comprises:
partitioning data into buckets; and
modeling the data distribution within each bucket as a linear spline;
wherein the partitioning includes:
a growth phase wherein the data space is recursively split into finer partitions; and
a pruning phase wherein some buckets are merged into bigger buckets.