Online trained object property estimator
AssigneeThroughPuter, Inc.
InventorMark Henrik Sandstrom
Life of this patent
Estimating the term…
Abstract
This disclosure describes systems and methods for using an estimator to produce values for dependent variables of streaming objects based on values of independent variables of the objects. The systems and methods may include continuously tuning the estimator based on any objects received with pre-populated values for the dependent variables.
Claims21
A method for estimating values of unknown features of a series of objects, the objects being represented as digital feature vectors, each respective digital feature vector including a plurality of X-variables having corresponding values populated on the respective digital feature vector before the estimating as a plurality of received X-variables, and, for at least a portion of the series of objects, a Y-variable having an unknown value prior to the estimating, the method comprising operations performed by a hierarchical estimator, the hierarchical estimator comprising hardware logic configured to perform at least a portion of the operations and/or software logic stored on a non-transitory digital medium and configured to perform, when executing via processing circuitry, at least a portion of the operations, the operations comprising:
- maintaining, by the hierarchical estimator on a non-transitory digital memory, an array of object models, wherein
- each model of the array of object models comprises at least a partial Y-variable value and a plurality of X-variable values corresponding to the respective at least a partial Y-variable value, and
- the array of object models is organized as a plurality of banks, with at least one given bank comprising a respective subset of the object models that are related by a respective common object classification of a plurality of classifications, wherein such classification corresponds to an identification of at least one defined value and/or value range for a qualitative or quantitative attribute that characterizes the respective subset of the object models of that classification according to the at least a partial Y-variable values of the respective subset of the object models;
- for each respective input object of at least a portion of the series of objects, in realtime,
- in a first stage of the hierarchical estimator, determining, for the respective input object, one or more candidate classifications from the plurality of classifications,
- in a second stage of the hierarchical estimator,
- for each of one or more given banks related to one of the one or more candidate classifications, identifying, from the object models in that given bank, a set of one or more closest matching models for the respective input object, and
- producing an estimated value of the Y-variable of the input object based at least in part on a Y-variable value of one or more models of the set of closest matching models for at least one of the banks associated with at least one of the one or more candidate classifications;
- and
- for given training objects of a series of training objects received interspersed in time with receipt of the input objects in the series of objects,
- updating a target object model from the array of object models, and corresponding to each given training object, in realtime.
A hierarchical estimator comprising:
- an array of object models stored in a non-transitory digital memory, wherein
- the array of object models is organized into a plurality of model banks, with one or more given model banks of the plurality of model banks comprising a respective subset of the object models that are related by a respective common object classification of a plurality of classifications, such respective common object classification corresponding to an identification of at least one defined value and/or value range for a qualitative or quantitative attribute that denotes the object models of that classification;
- first estimator stage logic, configured to identify, for an input object having at least one latent feature to be estimated, one or more candidate classifications from the plurality of classifications;
- second estimator stage logic, configured to apply the object models from one or more selected banks of the plurality of model banks to the input object, the selection of the one or more selected banks for the input object based on the one or more candidate classifications identified for the input object by the first estimator stage logic, wherein for at least one given bank of the one or more selected banks, applying the object models from that given bank comprises identifying a set of one or more closest matching models to the input object;
- output logic, configured to generate, based on the identification of the set of one or more closest matching models for at least one of the one or more selected banks, an estimated value for the at least one latent feature;
- and
- model update logic, configured to, in realtime and in response to receipt of a training object, update a target object model corresponding to the training object;
- wherein the hierarchical estimator is configured to perform interleaved processing of input objects having at least one feature to be estimated and training objects.
Dependencies are inferred from the claim text.