Granted Patent B2

US 10,409,909 · App. 14/569,104 · Granted Sep 10, 2019

Abstract

Claims (59)

1. A computer-implemented method for building a dictionary, the dictionary representing patterns associated with a behavior of an object, the method comprising:

receiving, via at least one processor, a stream of machine-readable symbols representing normalized data, each symbol being associated with a distinct cluster of vectors generated from the normalized data, the normalized data being obtained by normalizing data in each of a plurality of video frames, the plurality of video frames including a representation of an object;

performing, via the at least one processor, neural network based-linguistic analysis on the normalized data;

generating, via the at least one processor, a hierarchical learning model based on the neural network based-linguistic analysis;

determining, via the at least one processor, combinations of the symbols to identify as machine-readable words based on the hierarchical learning model, the hierarchical learning model having incremental learning levels for learning the machine readable words, each level indicating a length for the machine-readable words being identified at that level, the combinations of symbols to identify as machine-readable words at each level determined based on recurring combinations of symbols observed in the stream;

evaluating, via the at least one processor, statistics for the words identified at each level;

identifying, via the at least one processor, one or more of the words having statistical significance based on the evaluated statistics;

storing the identified one or more of the words having statistical significance and the statistics associated with each of the identified one or more of the words, the statistical significance being determined based on a measure of frequency at which the machine readable-words are observed in the stream of symbols;

determining, via the at least one processor, for an observation of at least a first one of the machine-readable words in the stream, an unusualness score; and

publishing an alert regarding the observation of the first machine readable word, the alert indicating a change in pattern associated with the behavior of the object.

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

receiving a second stream of machine-readable symbols;

determining machine-readable words from combinations of the symbols in the second stream based on the hierarchical learning model; and

updating, via the hierarchical learning model, the statistics of previously identified words.

3. The computer-implemented method of claim 1 , further comprising, generating a feature model from the identified one or more of the words having statistical significance, wherein the feature model includes generalizations of the identified one or more of the words having statistical significance.

4. The computer-implemented method of claim 3 , wherein generating the feature model comprises:

identifying features associated with the cluster of vectors associated with each symbol; and

generalizing the symbols based on the features.

5. The computer-implemented method of claim 3 , further comprising, evaluating statistics of the generalizations.

6. A non-transitory computer-readable storage medium storing instructions, which, when executed on a processor, performs an operation for building a dictionary of machine-readable words from combinations of machine-readable symbols, the dictionary representing patterns associated with a behavior of an object, the instructions comprising instructions to:

receive, via a processor, a stream of machine-readable symbols representing normalized data, each symbol associated with a distinct cluster of vectors generated from the normalized data, the normalized data being obtained by normalizing data in each of a plurality of video frames, the plurality of video frames including a representation of an object;

perform neural network based-linguistic analysis on the normalized data;

generate a hierarchical learning model based on the neural network based-linguistic analysis;

determine combinations of the symbols to identify as machine-readable words based on the hierarchical learning model, the hierarchical learning model having incremental learning levels for learning the machine readable words, each level indicating a length for the machine-readable words being identified at that level, the combinations of symbols to identify as machine-readable words at each level being determined based on recurring combinations of symbols observed in the stream;

evaluate statistics for the words identified at each level;

identify one or more of the words having statistical significance based on the evaluated statistics;

store the identified one or more of the words having statistical significance and the statistics associated with each of the identified one or more of the words, the statistical significance being determined based on a measure of frequency at which the machine readable-words are observed in the stream of symbols;

determine, for an observation of at least a first one of the machine-readable words in the stream, an unusualness score; and

publish an alert regarding the observation of the first machine readable word, the alert indicating a change in pattern associated with the behavior of the object.

7. The non-transitory computer-readable storage medium of claim 6 , wherein the instructions further comprises instructions to:

receive a second stream of machine-readable symbols;

determine machine-readable words from combinations of the symbols in the second stream based on the hierarchical learning model; and

update, via the hierarchical learning model, the statistics of previously identified words.

8. The non-transitory computer-readable storage medium of claim 6 , wherein the instructions further comprises instructions to, generate a feature model from the identified one or more of the words having statistical significance, wherein the feature model includes generalizations of the identified one or more of the words having statistical significance.

9. The non-transitory computer-readable storage medium of claim 8 , wherein generating the feature model comprises instructions to:

identify features associated with the cluster of vectors associated with each symbol; and

generalize the symbols based on the features.

10. The non-transitory computer-readable storage medium of claim 8 , wherein the instructions further comprises instructions to, evaluate statistics of the generalizations.

11. A system, comprising:

a processor; and

a memory storing one or more application programs configured to perform an operation for building a dictionary of machine-readable words from combinations of machine-readable symbols, the dictionary representing patterns associated with a behavior of an object, the operation comprising:

receiving, via a processor, a stream of machine-readable symbols representing normalized data, each symbol being associated with a distinct cluster of vectors generated from the normalized data, the normalized data being obtained by normalizing data in each of a plurality of video frames, the plurality of video frames including a representation of an object,

performing neural network based-linguistic analysis on the normalized data;

generating a hierarchical learning model based on the neural network based-linguistic analysis;

determining combinations of the symbols to identify as machine-readable words based on the hierarchical learning model, the hierarchical learning model having incremental learning levels for learning the machine readable words, each level indicating a length for the machine-readable words being identified at that level, and the combinations of symbols to identify as machine-readable words at each level being determined based on recurring combinations of symbols observed in the stream,

evaluating statistics for the words identified at each level,

identifying one or more of the words having statistical significance based on the evaluated statistics,

storing the identified one or more of the words having statistical significance and the statistics associated with each of the identified one or more of the words, the statistical significance being determined based on a measure of frequency at which the machine readable-words are observed in the stream of symbols,

determining, for an observation of at least a first one of the machine-readable words in the stream, an unusualness score, and

publishing an alert regarding the observation of the first machine readable word, the alert indicating a change in a pattern associated with the behavior of the object.

12. The system of claim 11 , wherein the operation further comprises:

receiving a second stream of machine-readable symbols;

determining machine-readable words from combinations of the symbols in the second stream based on the hierarchical learning model; and

updating, via the hierarchical learning model, the statistics of previously identified words.

13. The system of claim 11 , wherein the operation further comprises, generating a feature model from the identified one or more of the words having statistical significance, wherein the feature model includes generalizations of the identified one or more of the words having statistical significance.

14. The system of claim 13 , wherein generating the feature model comprises:

identifying features associated with the cluster of vectors associated with each symbol; and

generalizing the symbols based on the features.

15. The system of claim 13 , wherein the operation further comprises, evaluating statistics of the generalizations.

Assignments (69)