Granted Patent B2

US 9,996,502 · App. 15/589,922 · Granted Jun 12, 2018

Inventors: Rory Riggs (New York, NY); James Breitmeyer (New York, NY); Vin Harng Chew (London, GB); Daniel Goldman (New York, NY); Sean Sandys (Seattle, WA); Christopher Silkworth (Brooklyn, NY)

Assignee: Locus LP

G06F17/142G06F17/3053G06F17/30289G06F17/30312G06F17/30601G06F17/18G06F17/30G06Q40/06

Abstract

Claims (86)

1. A computational method for constructing a systems database in at least 10-dimensional space for real-time prediction of categorizations and interactions among elements in a functional system corresponding to data entities, the method comprising:

electronically storing a set of data entities in a logically structured table in first a database the data entities corresponding to elements of a functional system, wherein the functional system comprises a group of elements ordered by their functional roles in a process converting inputs to outputs;

electronically assigning one or more functional attributes to an element corresponding to a data entity in a logical data model structured as table in the first database that comprises at least two fields ordered by a set of interrelationships among at least two elements in the underlying functional system, the interrelationships corresponding to functional properties of a process converting a set of input elements to a set of output elements;

providing an electronic linking table between data tables containing the data entities and the functional attributes;

wherein the first database represents the group of elements subdivided into ordered sets of fields, a plurality of which have a defined logical data model in the first database with respect to functional attributes that are used to characterize the plurality of elements;

creating a systems database based on the first database by assigning a plurality of systems coordinates comprising an m-dimensional array of at least 10 -dimensional tensors to the data entities, wherein a plurality of entries in the array are based on the functional attributes of the elements and correspond to the functional properties;

algorithmically determining a reference distribution, wherein the reference distribution comprises the proportional allocation of elements into a first finite set of categories C=c 1,2 . . .p ;

using a statistical test T to assess the relative allocation of a set of data entities according to the reference distributions;

selecting an instance of a target distribution, wherein the target distribution comprises an algorithmic proportional assignment of data entities into a second finite set of categories;

electronically executing a statistical test T′ to assess the relative allocation in functional space of a set of data entities according to the target distribution;

selecting a set S=s 1, 2 . . . k of size k and dimension ≤M of at least 10-dimensional tensors defined by their functional distances, wherein functional distance is a measure of the relative remoteness of data entities in functional space;

electronically computing the difference between S and the remaining set of data entities L resulting in a matrix of dimension ≥10 (m−k), wherein the set of data entities in L are more functionally related than an arbitrary sample of data entities in S, as determined by a test of statistical significance;

electronically processing the matrix by a computerized machine learning algorithm to construct a probability space where a functional location of a tensor maps to a location with a corresponding probability for a plurality of categorizations;

electronically calculating a real-time prediction by a computationally efficient means using the electronic linking table, to with a given probability, where a data entity will be placed into a category; and

electronically storing the resulting prediction in the systems database in association with the data entity and the category.

2. The method of claim 1 , further comprising:

electronically computing the difference between S and the remaining set of data entities L resulting in a matrix of dimension ≥10(m−k), wherein the set of data entities in L are more functionally related than an arbitrary sample of data entities in S, as determined by a test of statistical significance;

electronically processing the matrix by a computerized machine learning algorithm to construct a probability space where a functional location of a tensor maps to a location with a corresponding probability for a plurality of categorizations;

assigning a scoring matrix of dimensionality ≤m×n comprising a set of weights associated with the m-dimensional array of n-dimensional tensors;

wherein the set of weights modifies the allocation of a variable across n-dimensional space so as to adjust the distribution, as determined by the statistical test; and

periodically rebalancing the set of weights associated with the tensors based on changes in the functional system;

wherein n is greater than or equal to 10.

3. The method of claim 2 , further comprising:

receiving the scoring matrix;

adding a k-dimensional set of n-dimensional tensors to the m-dimensional set of data entities;

using a machine learning technique to determine the new set of scores based on the locations of the tensors, the statistical test T′, and the target weight; and

outputting a matrix of dimension ≤(m+k)=n;

wherein the entries of the matrix comprise updated scores of the tensors and dimensions.

4. The method of claim 2 , further comprising:

receiving the scoring matrix;

subtracting a set of size k of n-dimensional tensors from the m-dimensional set of data entities;

using a machine learning technique to determine the new set of scores based on the location of the tensors, the statistical test T ′, and the target weight;

outputting a matrix of dimension ≥(m−k)×n;

wherein the entries of the matrix comprise updated scores of the tensors and dimensions.

5. The method of claim 4 , further comprising:

using a statistical measure of relatedness on the matrix to determine correspondence among functional and non-functional attributes in the matrix, thereby increasing the analytical performance compared to a non-filtered test on L.

6. The method of claim 2 , further comprising:

using the matrix representation of that coordinate space to predict, with a given probability, where a data entity will be placed into a category c; and

outputting an updated scoring matrix of dimension m′×n′.

7. The method of claim 2 , further comprising:

using the scoring matrix as an input to a machine learning technique to construct a coordinate space where tensors' functional locations form clusters based a plurality of categories;

using the matrix representation of that coordinate space to predict, with a given probability, where a data entity will be placed into a category; and

outputting an updated scoring matrix of dimension m″=n″.

8. The method of claim 1 , wherein:

the plurality of systems coordinates provide a basis for distinguishing a plurality of the specific roles of the elements in the underlying functional system, comparing the roles of a plurality of the elements, tracking the relative and absolute temporal evolution of the elements over time, or comparing the functional system to a second functional system.

9. A computational system constructing a systems database in at least 10-dimensional space for real-time prediction of categorizations and interactions among elements in a functional system corresponding to data entities, the system comprising a computerized processor configured for:

electronically storing a set of data entities in a logically structured table in first a database, the data entities corresponding to elements of a functional system, wherein the functional system comprises a group of elements ordered by their functional roles in a process converting inputs to outputs;

electronically assigning one or more functional attributes to an element corresponding to a data entity in a logical data model structured as table in the first database that comprises at least two fields ordered by a set of interrelationships among at least two elements in the underlying functional system, the interrelationships corresponding to functional properties of a process converting a set of input elements to a set of output elements;

providing an electronic linking table between data tables containing the data entities and the functional attributes;

wherein the first database represents the group of elements subdivided into ordered sets of fields, a plurality of which have a defined logical data model in the first database with respect to functional attributes that are used to characterize the plurality of elements;

creating a systems database based on the first database by assigning a plurality of systems coordinates comprising an m-dimensional array of at least 10-dimensional tensors to the data entities, wherein a plurality of entries in the array are based on the functional attributes of the elements and correspond to the functional properties;

algorithmically determining a reference distribution, wherein the reference distribution comprises the proportional allocation of elements into a first finite set of categories C=c 1, 2 . . . p ; using a statistical test T to assess the relative allocation of a set of data entities according to the reference distributions;

selecting an instance of a target distribution, wherein the target distribution comprises an algorithmic proportional assignment of data entities into a second finite set of categories;

electronically executing a statistical test T′ to assess the relative allocation in functional space of a set of data entities according to the target distribution;

selecting a set S=s 1, 2 . . . k of size k and dimension ≤m of at least 10-dimensional tensors defined by their functional distances, wherein functional distance is a measure of the relative remoteness of data entities in functional space;

electronically computing the difference between S and the remaining set of data entities L resulting in a matrix of dimension ≥10 (m−k), wherein the set of data entities in L are more functionally related than an arbitrary sample of data entities in S, as determined by a test of statistical significance;

electronically processing the matrix by a computerized machine learning algorithm to construct a probability space where a functional location of a tensor maps to a location with a corresponding probability for a plurality of categorizations; electronically calculating a real-time prediction by a computationally efficient means using the electronic linking table, to with a given probability, where a data entity will be placed into a category; and

electronically storing the resulting prediction in the systems database in association with the data entity and the category.

10. The system of claim 9 , further comprising:

assigning a scoring matrix of dimensionality ≤m×n comprising a set of weights associated with the m-dimensional array of n-dimensional tensors;

wherein the set of weights modifies the allocation of a variable across n-dimensional space so as to adjust the distribution, as determined by the statistical test; and

periodically rebalancing the set of weights associated with the tensors based on changes in the functional system.

11. The system of claim 10 , further comprising:

receiving the scoring matrix;

adding a k-dimensional set of n-dimensional tensors to the m-dimensional set of data entities;

using a machine learning technique to determine the new set of scores based on the locations of the tensors, the statistical test T′, and the target weight; and

outputting a matrix of dimension ≤(m+k)×n;

wherein the entries of the matrix comprise updated scores of the tensors and dimensions.

12. The system of claim 10 , further comprising:

receiving the scoring matrix;

subtracting a set of size k of n-dimensional tensors from the m-dimensional set of data entities;

using a machine learning technique to determine the new set of scores based on the location of the tensors, the statistical test T′, and the target weight;

outputting a matrix of dimension ≥(m−k)×n;

wherein the entries of the matrix comprise updated scores of the tensors and dimensions.

13. The system of claim 12 , further comprising:

using a statistical measure of relatedness on the matrix to determine correspondence among functional and non-functional attributes in the matrix, thereby increasing the analytical performance compared to a non-filtered test on L.

14. The system of claim 10 , further comprising:

using the scoring matrix as an input to a machine learning technique to construct a probability space where a functional location of a tensor maps to a location with a corresponding probability for a plurality of categorizations;

using the matrix representation of that coordinate space to predict, with a given probability, where a data entity will be placed into a category c; and

outputting an updated scoring matrix of dimension m′×n′.

15. The system of claim 10 , further comprising:

using the scoring matrix as an input to a machine learning technique to construct a coordinate space where tensors' functional locations form clusters based a plurality of categories;

using the matrix representation of that coordinate space to predict, with a given probability, where a data entity will be placed into a category; and

outputting an updated scoring matrix of dimension m″×n″.

16. The system of claim 9 , wherein:

the plurality of systems coordinates provide a basis for distinguishing a plurality of the specific roles of the elements in the underlying functional system, comparing the roles of a plurality of the elements, tracking the relative and absolute temporal evolution of the elements over time, or comparing the functional system to a second functional system.

Assignments (1)

Continuity (10)