Systems and methods for finding nearest neighbors
Systems and methods for finding nearest neighbors. In some embodiments, the system includes a processing circuit. The processing circuit may be configured to perform a method, the method including: selecting, based on a first query vector, a selected method, the selected method being a nearest neighbor selection method; and performing the selected method to select a first nearest neighbor from a data set, based on the first query vector.
1 . A system, comprising:
a processor, the processor being configured to perform a method, the method comprising:
receiving, by the processor, a first query vector;
determining that the first query vector belongs to a first class of query vectors having a first computational characteristic that is different from a second computational characteristic of query vectors of a second class;
selecting, by a machine learning model comprising a classifier, based on the first class and from between at least a first method providing a first performance and a second method providing a second performance that is different from the first performance, the first method as a selected method based on the first performance being different from the second performance, the selected method being a nearest neighbor selection method;
performing the selected method to select a first nearest neighbor from a data set, based on the first query vector;
performing the second method, different from the selected method, to select a second nearest neighbor from the data set, based on the first query vector; and
training the classifier based on a difference between the first nearest neighbor and the second nearest neighbor,
wherein the selecting of the selected method comprises selecting, by the classifier, the selected method.
2 . The system of claim 1 , wherein the classifier comprises a neural network, and the training comprises performing back-propagation.
3 . The system of claim 1 , wherein the method further comprises
performing a third method to select a third nearest neighbor from the data set, based on a second query vector;
performing a fourth method, different from the third method, to select a fourth nearest neighbor from the data set, based on the second query vector; and
training the classifier based on a difference between the third nearest neighbor and the fourth nearest neighbor.
4 . The system of claim 3 , further comprising processing a first number of query vectors between the processing of the first query vector and the processing of the second query vector.
5 . The system of claim 4 , wherein the first number is a constant.
6 . The system of claim 4 , wherein the first number is generated by a pseudorandom number generator.
7 . The system of claim 4 , wherein the first number is based on a measure of a difference between an output of the selected method and an output of the second method.
8 . The system of claim 7 , wherein the measure of the difference comprises a Euclidean distance.
9 . The system of claim 7 , wherein the measure of the difference comprises a Manhattan distance.
10 . The system of claim 7 , wherein the measure of the difference comprises a cosine similarity.
11 . The system of claim 1 , wherein the selecting of the selected method comprises selecting the selected method based on a principal component analysis conversion error rate.
12 . The system of claim 1 , wherein the selecting of the selected method comprises selecting the selected method based on a linear discriminant analysis.
13 . A method, comprising:
receiving a first query vector;
determining that the first query vector belongs to a first class of query vectors having a first computational characteristic that is different from a second computational characteristic of query vectors of a second class;
selecting, by a machine learning model comprising a classifier, based on the first class and from between at least a first method providing a first energy usage and a second method providing a second energy usage that is different from the first energy usage, the first method as a selected method based on the first energy usage being different from the second energy usage, the selected method being a nearest neighbor selection method;
performing the selected method to select a first nearest neighbor from a data set, based on the first query vector;
performing the second method, different from the selected method, to select a second nearest neighbor from the data set, based on the first query vector; and
training the classifier based on a difference between the first nearest neighbor and the second nearest neighbor,
wherein the selecting of the selected method comprises selecting, by the classifier, the selected method.
14 . A system, comprising:
a means for processing, the means for processing being configured to perform a method, the method comprising:
receiving, by the means for processing, a first query vector;
determining that the first query vector belongs to a first class of query vectors having a first computational characteristic that is different from a second computational characteristic of query vectors of a second class;
selecting, by a machine learning model comprising a classifier, based on the first class and from between at least a first method providing a first accuracy and a second method providing a second accuracy that is different from the first accuracy, the first method as a selected method based on the first accuracy being different from the second accuracy, the selected method being a nearest neighbor selection method;
performing the selected method to select a first nearest neighbor from a data set, based on the first query vector;
performing the second method, different from the selected method, to select a second nearest neighbor from the data set, based on the first query vector; and
training the classifier based on a difference between the first nearest neighbor and the second nearest neighbor,
wherein the selecting of the selected method comprises selecting, by the classifier, the selected method.