Granted Patent B2

US 12,681,993 · App. 18/643,602 · Granted Jul 14, 2026

Inventors: Luis F. Stevens (San Jose, CA); John Skolfield (San Jose, CA); Vince Schiavone (Berwyn, PA); Steve Weissinger (San Jose, CA); George Tretyakov (Campbell, CA); Ryan Mammina (San Jose, CA); Mo Malakiman (San Jose, CA)

Assignee: Target Brands, Inc.

G06F16/90335G06F16/24568G06F40/237G06F40/284G06F40/30

Abstract

Claims (59)

1 . A method of classifying data streams in real-time, the method comprising:

receiving a data stream that includes a plurality of data packets;

segmenting each data packet of the plurality of data packets into a plurality of data segments;

calculating a L-2 norm for each data segment of the plurality of data segments for each data packet of the plurality of data packets;

based on the L-2 norm for each data segment of the plurality of data segments for each data packet of the plurality of data packets, clustering the plurality of data segments into a plurality of data clusters, wherein each of the plurality of data clusters comprises one or more data segments of the plurality of data segments;

aggregating the plurality of data clusters associated with each data packet of the plurality of data packets into a respective aggregated data cluster, wherein the aggregating is performed by multi-threaded aggregating processes;

using a lockless hash-map, computing a plurality of min-distance vectors associated with each respective aggregated data cluster associated with each data packet of the plurality of data packets in parallel and asynchronously, wherein each min-distance vector of the plurality of min-distance vectors corresponds to each data packet of the plurality of data packets;

sorting each respective min-distance vector in the plurality of min-distance vectors to compute a plurality of minimum distances for each respective data packet in the plurality of data packets, wherein the sorting is performed by multi-threaded sorting processes; and

outputting, for each respective data packet of the plurality of data packets, the plurality of minimum distances that corresponds to the respective data packet.

2 . The method of claim 1 , further comprising:

while segmenting the respective data packet of the plurality of data packets, assigning a unique non-repeating packet identifier to the respective data packet, and annotating each respective data segment corresponding to the respective data packet with the unique non-repeating packet identifier; and

while aggregating the plurality of data clusters, using the unique non-repeating packet identifier that the respective data segment is annotated with as a key for the lockless hash-map.

3 . The method of claim 1 , wherein a plurality of classifying engines is used in clustering the plurality of data segments.

4 . The method of claim 3 , further comprising: selecting a number of classifying engines in the plurality of classifying engines based on the received data stream and using a subset of the plurality of classifying engines to cluster the plurality of data segments.

5 . The method of claim 3 , wherein each classifying engine of the plurality of classifying engines is a Linear Algebra engine.

6 . The method of claim 2 , wherein clustering the plurality of data segments further includes:

training a classification model to compute centroids for K clusters in N dimensions, wherein K is a predetermined number of classification categories and N is a predetermined data-dimensionality for the data stream; and

calculating the L2-norm for each data segment of the plurality of data segments using the computed centroids or a predetermined list of centroids.

7 . The method of claim 6 , wherein outputting the plurality of minimum distances further includes tabulating the plurality of minimum distances along with information related to the computed centroid of the clusters and the packet identifiers for the plurality of data packets.

8 . The method of claim 1 , wherein sorting to compute the plurality of minimum distances further includes identifying a respective min-distance vector of the plurality of min-distance vectors that corresponds to a respective data packet based on a packet identifier corresponding to the respective data packet and sorting the respective min-distance vector.

9 . The method of claim 1 , wherein aggregating the plurality of data clusters and sorting to compute the plurality of minimum distances are implemented as multi-threaded processes.

10 . A system for classifying data streams in real-time, the system comprising:

a computing system with one or more processors and memory, the memory storing instructions that when executed by the one or more processors cause the computing system to:

receive a data stream that includes a plurality of data packets;

segment each data packet of the plurality of data packets into a plurality of data segments;

store the plurality of data segments in a first memory buffer, while continuing to receive data streams;

cluster the plurality of data segments into a plurality of data clusters, wherein the clustering includes:

read the plurality of data segments from the first memory buffer; and

calculate a L-2 norm for each data segment of the plurality of data segments, wherein each data cluster of the plurality of data clusters comprises one or more data segments of the plurality of data segments;

store the plurality of data clusters to a second memory buffer;

aggregate the plurality of data clusters, using a lockless hash map, to compute a plurality of min-distance vectors in parallel and asynchronously, including by reading one or more data clusters of the plurality of data clusters from the second memory buffer, wherein the aggregating is performed by multi-threaded aggregating processes;

store the plurality of min-distance vectors to a third memory buffer, wherein each respective min-distance vector in the plurality of min-distance vectors corresponds to a data packet in the plurality of data packets, wherein the sorting is performed by multi-threaded sorting processes;

sort each respective min-distance vector in the plurality of min-distance vectors to compute a plurality of minimum distances for each respective data packet of the plurality of data packets, including by reading the plurality of min-distance vectors from the third memory buffer; and

output, for each respective data packet of the plurality of data packets, the plurality of minimum distances that corresponds to the respective data packet.

11 . The system of claim 10 , wherein the instructions when executed by the one or more processors further cause the computing system to:

while segmenting the respective data packet of the plurality of data packets, assign a unique non-repeating packet identifier to the respective data packet, and annotating each respective data segment corresponding to the respective data packet with the unique non-repeating packet identifier; and

while aggregating the plurality of data clusters, use the unique non-repeating packet identifier that the respective data segment is annotated with as a key for the lockless hash-map.

12 . The system of claim 10 , wherein a plurality of classifying engines is used in clustering the plurality of data segments.

13 . The system of claim 12 , wherein the instructions when executed by the one or more processors further cause the computing system to:

select a number of classifying engines in the plurality of classifying engines based on the received data stream and using a subset of the plurality of classifying engines to cluster the plurality of data segments.

14 . The system of claim 12 , wherein each classifying engine of the plurality of classifying engines is a Linear Algebra based engine.

15 . The system of claim 11 , wherein to cluster the plurality of data segments further includes to:

train a classification model to compute centroids for K clusters in N dimensions, wherein K is a predetermined number of classification categories and N is a predetermined data-dimensionality for the data stream; and

calculate the L2-norm for each data segment of the plurality of data segments using the computed centroids or a predetermined list of centroids.

16 . The system of claim 15 , wherein to output the plurality of minimum distances further includes to tabulate the plurality of minimum distances along with information related to the computed centroid of the clusters and the packet identifiers for the plurality of data packets.

17 . The system of claim 10 , wherein to sort to compute the plurality of minimum distances further includes to identify a respective min-distance vector of the plurality of min-distance vectors that corresponds to a respective data packet based on a packet identifier corresponding to the respective data packet and sorting the respective min-distance vector.

18 . The system of claim 10 , wherein aggregating the plurality of data clusters and sorting to compute the plurality of minimum distances are implemented as multi-threaded processes.

19 . A non-transitory computer readable storage medium storing a set of instructions, which when executed by a computing system with one or more processors cause the computing system to:

receive a data stream that includes a plurality of data packets;

segment each data packet of the plurality of data packets into a plurality of data segments;

calculate a L-2 norm for each data segment of the plurality of data segments for each data packet of the plurality of data packets;

based on the L-2 norm for each data segment of the plurality of data segments for each data packet of the plurality of data packets, cluster the plurality of data segments into a plurality of data clusters, wherein each of the plurality of data clusters comprises one or more data segments of the plurality of data segments;

aggregate the plurality of data clusters associated with each data packet of the plurality of data packets into a respective aggregated data cluster, wherein the aggregating is performed by multi-threaded aggregating processes;

using a lockless hash-map, compute a plurality of min-distance vectors associated with each respective aggregated data cluster associated with each data packet of the plurality of data packets in parallel and asynchronously, wherein each min-distance vector of the plurality of min-distance vectors corresponds to each data packet of the plurality of data packets; and

sort each respective min-distance vector in the plurality of min-distance vectors to compute a plurality of minimum distances for each respective data packet in the plurality of data packets, wherein the sorting is performed by multi-threaded sorting processes; and

output, for each respective data packet of the plurality of data packets, the plurality of minimum distances that corresponds to the respective data packet.

20 . The system of claim 19 , wherein to cluster the plurality of data segments further includes to:

train a classification model to compute centroids for K clusters in N dimensions, wherein K is a predetermined number of classification categories and N is a predetermined data-dimensionality for the data stream; and

calculate the L2-norm for each data segment of the plurality of data segments using the computed centroids or a predetermined list of centroids.

Continuity (5)

References Cited (112)