Systems and methods for robust pattern classification
AssigneeQUALCOMM Incorporated
InventorHasib Ahmed Siddiqui
Life of this patent
Estimating the term…
Abstract
Certain embodiments relate to systems and methods for performing data discrimination using dimensionality reduction techniques. Particularly the Sparse Matrix Transform (SMT) is applied to more efficiently determine the Fischer Discrimination vector for a given dataset. Adjustments to the SMT facilitate more robust identification of the Fischer Discrimination vector in view of various resource constraints.
Claims23
A method, implemented on an electronic device, for generating physical sensor data classifiers, the method comprising:
- receiving a plurality of physical sensor data;
- identifying a projection vector, comprising:
- using a search algorithm comprising a metric function to identify correlated data represented by a sub-matrix of the plurality of sensor data, and
- calculating one or more eigenvalues associated with the metric function at least in part by transforming the plurality of physical sensor data using a sparse matrix transform (SMT),
- wherein transforming using the SMT comprises reducing off-diagonal elements of the sub-matrix, wherein the off-diagonal elements represent a level of correlation of the correlated data;
- and
- producing physical sensor data classifiers by projecting at least a portion of the physical sensor data upon the projection vector.
A mobile electronic device comprising:
- a memory, the memory configured to store a data set comprising physical sensor data;
- a processor configured to:
- receive a plurality of physical sensor data;
- identify a projection vector, wherein identifying comprises:
- using a search algorithm comprising a metric function to identify correlated data represented by a sub-matrix of the plurality of sensor data, and
- calculating one or more eigenvalues associated with the metric function at least in part by transforming the plurality of physical sensor data using a sparse matrix transform (SMT),
- wherein transforming using the SMT comprises reducing off-diagonal elements of the sub-matrix, wherein the off-diagonal elements represent a level of correlation of the correlated data;
- and
- produce physical sensor data classifiers by projecting at least a portion of the physical sensor data upon the projection vector.
A non-transitory, computer-readable medium, comprising instructions configured to cause a processor to implement a method to classify physical sensor data, the method comprising:
- receiving a plurality of physical sensor data;
- identifying a projection vector, comprising:
- using a search algorithm comprising a metric function to identify correlated data represented by a sub-matrix of the plurality of sensor data, and
- calculating one or more eigenvalues associated with the metric function at least in part by transforming the plurality of physical sensor data using a sparse matrix transform (SMT),
- wherein transforming using the SMT comprises reducing off-diagonal elements of the sub-matrix, wherein the off-diagonal elements represent a level of correlation of the correlated data;
- and
- producing physical sensor data classifiers by projecting at least a portion of the physical sensor data upon the projection vector.
A mobile electronic device comprising:
- means for receiving a plurality of physical sensor data;
- means for identifying a projection vector, wherein identifying the projection vector comprises:
- using a search algorithm comprising a metric function to identify correlated data represented by a sub-matrix of the plurality of sensor data, and
- calculating one or more eigenvalues associated with the metric function at least in part by transforming the plurality of physical sensor data using a sparse matrix transform (SMT),
- wherein transforming using the SMT comprises reducing off-diagonal elements of the sub-matrix, wherein the off-diagonal elements represent a level of correlation of the correlated data;
- and
- means for producing physical sensor data classifiers by projecting at least a portion of the physical sensor data upon the projection vector.
Dependencies are inferred from the claim text.