Technologies for refining stochastic similarity search candidates

1. A device comprising:

a memory that is column addressable and row addressable to allow both row-wise writes and column-wise reads;

circuitry connected to the memory, wherein the circuitry is to:

add a set of input data vectors to the memory as a set of binary dimensionally expanded vectors, including multiplying each input data vector with a projection matrix to project each input data vector from a first dimension space to a second expanded dimensional space, wherein the projection matrix is a binary sparse projection matrix of random binary values or a dense floating point matrix;

produce a search hash code from a search data vector, including multiplying the search data vector with the projection matrix;

identify a result set of the binary dimensionally expanded vectors as a function of a Hamming distance of each binary dimensionally expanded vector from the search hash code; and

determine, from the result set, a refined result set as a function of a similarity measure in an original input space of the input data vectors,

wherein the operations to add the set of input data vectors, produce the search hash code, and identify the result set comprise a stochastic associative search, and wherein the stochastic associative search includes row-wise writes of a block of one or more specific rows and column-wise reads of a block of one or more specific columns from the memory.

2. The device of claim 1 , wherein the circuitry is further to train an inverse map to convert the binary dimensionally expanded vectors and the search hash code to an original data format.

3. The device of claim 2 , wherein to determine a refined result set comprises to convert the binary dimensionally expanded vectors in the result set to vectors of floating point values.

4. The device of claim 2 , wherein to determine a refined result set comprises to convert the search hash code to a set of floating point values.

5. The device of claim 2 , wherein to determine a refined result set comprises to determine a Euclidean distance between each converted vector and the converted search hash code.

6. The device of claim 5 , wherein to determine the refined result set further comprises to identify a subset of the converted vectors having the lowest Euclidean distance from the converted search hash code as the refined result set.

7. The device of claim 1 , wherein to determine, from the result set, a refined result set comprises to determine a refined result set that is an order of magnitude smaller than the result set.

8. The device of claim 1 , wherein multiplying each input data vector with a projection matrix comprises multiplying input data vectors having data values indicative of features extracted from objects to be searched with the projection matrix.

9. The device of claim 8 , wherein multiplying each input data vector with a projection matrix comprises multiplying input data vectors having data values indicative of features extracted from images to be searched with the projection matrix.

10. The device of claim 1 , wherein the circuitry is further to sort the binary dimensionally expanded vectors as a function of the Hamming distance of each binary dimensionally expanded vector from the search hash code.

11. The device of claim 1 , wherein the memory has a three dimensional cross point architecture.

12. A method comprising:

adding, by a device having a memory that is column addressable and row addressable to allow both row-wise writes and column-wise reads, a set of input data vectors to the memory as a set of binary dimensionally expanded vectors, including multiplying each input data vector with a projection matrix to project each input data vector from a first dimension space to a second expanded dimensional space, wherein the projection matrix is a binary sparse projection matrix of random binary values or a dense floating point matrix;

producing, by the device, a search hash code from a search data vector, including multiplying the search data vector with the projection matrix;

identifying, by the device, a result set of the binary dimensionally expanded vectors as a function of a Hamming distance of each binary dimensionally expanded vector from the search hash code; and

determining, by the device and from the result set, a refined result set as a function of a similarity measure in an original input space of the input data vectors,

wherein the operations of adding the set of input data vectors, producing the search hash code, and identifying the result set comprise a stochastic associative search, and wherein the stochastic associative search includes row-wise writes of a block of one or more specific rows and column-wise reads of a block of one or more specific columns from the memory.

13. The method of claim 12 , further comprising training, by the device, an inverse map to convert the binary dimensionally expanded vectors and the search hash code to an original data format.

14. A system comprising:

a processor;

a memory that is column addressable and row addressable to allow both row-wise writes and column-wise reads;

circuitry connected to the memory, wherein the circuitry is to:

add a set of input data vectors to the memory as a set of binary dimensionally expanded vectors, including multiplying each input data vector with a projection matrix to project each input data vector from a first dimension space to a second expanded dimensional space, wherein the projection matrix is a binary sparse projection matrix of random binary values or a dense floating point matrix;

produce a search hash code from a search data vector, including multiplying the search data vector with the projection matrix;

identify a result set of the binary dimensionally expanded vectors as a function of a Hamming distance of each binary dimensionally expanded vector from the search hash code; and

determine, from the result set, a refined result set as a function of a similarity measure in an original input space of the input data vectors,

wherein the operations to add the set of input data vectors, produce the search hash code, and identify the result set comprise a stochastic associative search, and wherein the stochastic associative search includes row-wise writes of a block of one or more specific rows and column-wise reads of a block of one or more specific columns from the memory.

15. The system of claim 14 , wherein the circuitry is in a data storage device.

16. The system of claim 14 , wherein the circuitry is in a memory device.

17. The system of claim 14 , wherein the circuitry is further to train an inverse map to convert the binary dimensionally expanded vectors and the search hash code to an original data format.

18. The system of claim 17 , wherein to determine a refined result set comprises to convert the binary dimensionally expanded vectors in the result set to vectors of floating point values.

19. One or more non-transitory machine-readable storage media comprising a plurality of instructions stored thereon that, in response to being executed, cause a device having a memory that is column addressable and row addressable to allow both row-wise writes and column-wise reads to:

add a set of input data vectors to the memory as a set of binary dimensionally expanded vectors, including multiplying each input data vector with a projection matrix to project each input data vector from a first dimension space to a second expanded dimensional space, wherein the projection matrix is a binary sparse projection matrix of random binary values or a dense floating point matrix;

produce a search hash code from a search data vector, including multiplying the search data vector with the projection matrix;

identify a result set of the binary dimensionally expanded vectors as a function of a Hamming distance of each binary dimensionally expanded vector from the search hash code; and

determine, from the result set, a refined result set as a function of a similarity measure in an original input space of the input data vectors,

wherein the operations to add the set of input data vectors, produce the search hash code, and identify the result set comprise a stochastic associative search, and wherein the stochastic associative search includes row-wise writes of a block of one or more specific rows and column-wise reads of a block of one or more specific columns from the memory.