Perceptual associative memory for a neuro-linguistic behavior recognition system

• US 10,409,910 B2
• Filed: 12/12/2014
• Issued: 09/10/2019
• Est. Priority Date: 12/12/2014
• Status: Active Grant

• Current Assignee: Intellective Ai, Inc.
• Original Assignee: Omni AI, Inc.
• Inventors: Seow, Ming-Jung, Xu, Gang, Yang, Tao, Cobb, Wesley Kenneth
• Primary Examiner(s): Adesanya, Olujimi A
• Application Number:
  US14/569,161
• Publication Number:
  US 20160170961A1
• Time in Patent Office: 1,733 Days
• Field of Search: 704 9, 704 10, 704232, 704243, 704245
• CPC Class Codes: G06F 40/242 Dictionaries
• CPC Class Codes: G06F 40/284 Lexical analysis, e.g. tokenisation or collocates
• CPC Class Codes: G06F 40/40 Processing or translation of natural language natural language analysis G06F40/20; semantic analysis G06F40/30
• CPC Class Codes: G06N 3/0409 Adaptive resonance theory [ART] networks
• CPC Class Codes: G06N 3/042 Knowledge-based neural networks; Logical representations of neural networks
• CPC Class Codes: G06N 3/047 Probabilistic or stochastic networks
• CPC Class Codes: G06N 3/088 Non-supervised learning, e.g. competitive learning
• CPC Class Codes: G06N 5/022 Knowledge engineering; Knowledge acquisition

Abstract

Techniques are disclosed for generating a syntax for a neuro-linguistic model of input data obtained from one or more sources. A stream of words of a dictionary built from a sequence of symbols are received. The symbols are generated from an ordered stream of normalized vectors generated from input data. Statistics for combinations of words co-occurring in the stream are evaluated. The statistics includes a frequency upon which the combinations of words co-occur. A model of combinations of words based on the evaluated statistics is updated. The model identifies statistically relevant words. A connected graph is generated. Each node in the connected graph represents one of the words in the stream. Edges connecting the nodes represent a probabilistic relationship between words in the stream. Phrases are identified based on the connected graph.

Claims

1. A computer-implemented method for analyzing behavior of an object, the method comprising:

receiving a plurality of video frames from a video source, the plurality of video frames including a representation of an object;

normalizing, via at least one processor, data in each video frame from the received plurality of video frames to obtain normalized data for the plurality of video frames;

performing, via the at least one processor, neuro-linguistic analysis on the normalized data;

generating, via the at least one processor, a syntax for a neuro-linguistic model based on the neuro-linguistic analysis, the syntax including a stable model of phrases, wherein generating the syntax includes;

generating, via the at least one processor, machine-readable symbols from the normalized data, each machine-readable symbol being associated with a distinct cluster of the normalized data, and building a dictionary of machine-readable words based on an observed-sequence of the machine-readable symbols;

evaluating, via the at least one processor, statistics for combinations of machine-readable words co-occurring in a stream of machine-readable words, the statistics including a frequency at which the combinations of words co-occur;

updating, via the at least one processor, a model of combinations of machine-readable words based on the evaluated statistics, the model identifying statistically relevant observations of machine-readable words co-occurring in the input stream of machine-readable symbols; and

generating, via the at least one processor, a connected graph, each node in the connected graph representing one of the words in the stream, and edges connecting the nodes representing a probabilistic measure of co-occurrence of pairs of statistically relevant words in the stream connected by an edge of the graph;

identifying, via the at least one processor, instances of at least one machine-readable phrase based on the connected graph;

determining, via the at least one processor, an unusualness score for an observation of at least a first one of the machine-readable phrases identified in the connected graph; and

publishing, via the at least one processor, an alert regarding the observation of the first machine-readable phrase, the alert indicating an anomaly in the behavior of the object.

2. The computer-implemented method of claim 1, wherein the statistics include a statistical significance score indicating a frequency that two words co-occur with one another.

3. The computer-implemented method of claim 2, further comprising:

identifying one or more phrases, from the stable model of phrases, that include co-occurring words having a statistical significance score that exceeds a specified threshold; and

outputting the identified phrases.

4. The computer-implemented method of claim 1, wherein the edges are weighted based on a statistical significance score between each node.

5. The computer-implemented method of claim 1, further comprising:

receiving a second stream of machine-readable words; and

updating the statistics of the co-occurring words based on the second stream of machine-readable words.

6. The computer-implemented method of claim 5, further comprising:

decreasing a statistical significance score of co-occurring words that are less frequently observed over time.

7. The computer-implemented method of claim 5, further comprising:

increasing a statistical significance score of co-occurring words that are more frequently observed over time.

8. A non-transitory computer-readable storage medium storing instructions, which, when executed on a processor, performs an operation for analyzing behavior of an object, the instructions comprising instructions to:

receive a plurality of video frames including a representation of an object;

normalize data in each video frame from the received plurality of video frames to obtain normalized data for the plurality of video frames;

perform neuro-linguistic analysis on the normalized data;

generate a syntax for a neuro-linguistic model based on the neuro-linguistic analysis, the syntax including a stable model of phrases, wherein the instructions to generate the syntax includes instructions to;

generate machine-readable symbols from the normalized data, each machine-readable symbol being associated with a distinct cluster of the normalized data, and build a dictionary of machine-readable words based on an observed-sequence of the machine-readable symbols;

evaluate statistics for combinations of machine-readable words co-occurring in a stream of machine-readable words, the statistics including a frequency at which the combinations of words co-occur;

update a model of combinations of machine-readable words based on the evaluated statistics, the model identifying statistically relevant observations of machine-readable words co-occurring in the input stream of machine-readable symbols; and

generate a connected graph, each node in the connected graph representing one of the words in the stream, and edges connecting the nodes representing a probabilistic measure of co-occurrence of pairs of statistically relevant words in the stream connected by an edge of the graph;

identify instances of at least one machine-readable phrase, from the stable model of phrases, based on the connected graph;

determine an unusualness score for an observation of at least a first one of the machine-readable phrases identified in the connected graph; and

publish an alert regarding the observation of the first machine-readable phrase, the alert indicating an anomaly in the behavior of the object.

9. The non-transitory computer-readable storage medium of claim 8, wherein the statistics include a statistical significance score indicating a frequency that two words co-occur with one another.

10. The non-transitory computer-readable storage medium of claim 9, wherein the instructions further comprises instructions to:

identify one or more phrases from the stable model of phrases that includes co occurring words having a statistical significance score that exceeds a specified threshold; and

output the identified phrases.

11. The non-transitory computer-readable storage medium of claim 8, wherein the edges are weighted based on a statistical significance score between each node.

12. The non-transitory computer-readable storage medium of claim 8, wherein the instructions further comprises instructions to:

receive a second stream of words; and

update the statistics of the co-occurring words based on the second stream of words.

13. The non-transitory computer-readable storage medium of claim 12, wherein the instructions further comprises instructions to:

decrease a statistical significance score of co-occurring words that are less frequently observed over time.

14. The non-transitory computer-readable storage medium of claim 12, wherein the instructions further comprises instructions to:

increase a statistical significance score of co-occurring words that are more frequently observed over time.

15. A system, comprising:

a processor; and

a memory storing one or more application programs configured to perform an operation for analyzing behavior of an object, the operation comprising;

receiving a plurality of video frames including a representation of an object;

normalizing data in each video frame from the received plurality of video frames to obtain normalized data for the plurality of video frames;

performing neuro-linguistic analysis on the normalized data;

generating a syntax for a neuro-linguistic model based on the neuro-linguistic analysis, the syntax including a stable model of phrases, wherein generating the syntax includes;

generating machine-readable symbols from the normalized data, each machine-readable symbol being associated with a distinct cluster of the normalized data, and building a dictionary of machine-readable words based on an observed-sequence of the machine-readable symbols;

evaluating statistics for combinations of machine-readable words co-occurring in a stream of machine-readable words, the statistics including a frequency at which the combinations of words co-occur,

updating a model of combinations of machine-readable words based on the evaluated statistics, the model identifying statistically relevant observations of machine-readable words co-occurring in the input stream of machine-readable symbols, and

generating a connected graph, each node in the connected graph representing one of the words in the stream, and edges connecting the nodes representing a probabilistic measure of co-occurrence of pairs of statistically relevant words in the stream connected by an edge of the graph,

identifying instances of at least one machine-readable phrase, from the stable model of phrases, based on the connected graph,

determining an unusualness score for an observation of at least a first one of the machine-readable phrases identified in the connected graph, and

issuing an alert regarding the observation of the first machine-readable phrase, the alert indicating an anomaly in the behavior of the object.

16. The system of claim 15, wherein the statistics include a statistical significance score indicating a frequency that two words co-occur with one another.

17. The system of claim 16, wherein the operation further comprises:

identifying one or more phrases, from the stable model of phrases, that include co-occurring words having a statistical significance score that exceeds a specified threshold; and

outputting the identified phrases.

18. The system of claim 15, wherein the operation further comprises:

receiving a second stream of machine-readable words; and

updating the statistics of the co-occurring words based on the second stream of machine-readable words.

19. The system of claim 18, wherein the operation further comprises:

decreasing a statistical significance score of co-occurring words that are less frequently observed over time.

20. The system of claim 18, wherein the operation further comprises:

increasing a statistical significance score of co-occurring words that are more frequently observed over time.