Methods and apparatus for visualization of machine learning malware detection models
Embodiments disclosed include methods and apparatus for visualization of data and models (e.g., machine learning models) used to monitor and/or detect malware to ensure data integrity and/or to prevent or detect potential attacks. Embodiments disclosed include receiving information associated with artifacts scored by one or more sources of classification (e.g., models, databases, repositories). The method includes receiving inputs indicating threshold values or criteria associated with a classification of maliciousness of an artifact and for selecting sample artifacts. The method further includes classifying and selecting the artifacts, based on the criteria, to define a sample set, and based on the sample set, generating a ground truth indication of classification of maliciousness for each sample artifact in the sample set. The method further includes using the ground truth indications to evaluate and display, via an interface, a representation of a performance of sources of classification and/or quality of data.
1 . One or more non-transitory processor readable media storing code representing instructions to be executed by one or more processors, the instructions comprising code to cause the one or more processors to:
receive information associated with a plurality of artifacts, the information associated with each artifact from the plurality of artifacts being based on a source of classification from a plurality of sources of classification;
receive a criterion for selecting sample artifacts from the plurality of artifacts;
select, based on the criterion, sample artifacts from the plurality of artifacts to define a sample set;
determine, based on the selecting, a ground truth indication of classification of maliciousness for each sample artifact in the sample set; and
display, via an interface and based on the ground truth indication of classification of maliciousness for one or more sample artifacts in the sample set, a representation of a performance of a source of classification from the plurality of sources of classification.
2 . The one or more non-transitory processor readable media of claim 1 , wherein the criterion indicates a minimum number of sources of classification from the plurality of sources of classification associated with a sample artifact from the plurality of artifacts.
3 . The one or more non-transitory processor readable media of claim 1 , wherein the criterion is a first criterion, the one or more non-transitory processor readable media further comprising code to cause the one or more processors to:
receive a second criterion for selecting sample artifacts from the plurality of artifacts, the second criterion indicating a minimum number of scores associated with an artifact, each score from the minimum number of scores being associated with a classification of maliciousness of that artifact from an output from at least one source of classification from the plurality of sources of classification, the code to cause the one or more processors to determine includes code to cause the one or more processors to determine the ground truth indication of classification of maliciousness for each sample artifact in the sample set based on the second criterion.
4 . The one or more non-transitory processor readable media of claim 1 , wherein each artifact from the plurality of artifacts includes at least one of a file, a uniform resource locator (URL) or a device.
5 . The one or more non-transitory processor readable media of claim 1 , wherein the plurality of sources of classification includes at least one machine learning (ML) model trained to classify potentially malicious artifacts.
6 . The one or more non-transitory processor readable media of claim 1 , wherein the source of classification includes a machine learning (ML) model trained to classify potentially malicious artifacts, the instructions further comprising code to cause the one or more processors to:
automatically retrain the ML model based on the representation of the performance of the source of classification.
7 . The one or more non-transitory processor readable media of claim 1 , further comprising instructions to cause the one or more processors to:
receive, via the interface, new information associated with an artifact from the plurality of artifacts; and
update, based on the new information, the ground truth indication of the classification of maliciousness of that artifact.
8 . The one or more non-transitory processor readable media of claim 1 , wherein the criterion is user-selectable.
9 . The one or more non-transitory processor readable media of claim 1 , further comprising instructions to cause the one or more processors to:
receive, via the interface, a threshold score associated with a classification of maliciousness for each artifact from the plurality of artifacts, the code to cause the one or more processors to determine including code to cause the one or more processors to determine the ground truth indication of classification of maliciousness for each sample artifact in the sample set based on the threshold score for that sample artifact.
10 . One or more non-transitory processor readable media storing code representing instructions to be executed by one or more processors, the instructions comprising code to cause the one or more processors to:
receive, at a first time, a plurality of scores associated with a classification of maliciousness of an artifact, each score from the plurality of scores being from a different source of classification from a plurality of sources of classification than remaining scores from the plurality of scores;
determine, based on the plurality of scores, a ground truth indication of classification of maliciousness for the artifact;
receive, at a second time after the first time, at least one updated score associated with the classification of maliciousness of the artifact, the at least one updated score being from at least one source of classification from the plurality of sources of classification;
update, based on the at least one updated score, the ground truth indication of classification of maliciousness for the artifact to produce an updated ground truth indication of classification of maliciousness for the artifact; and
display, via an interface and based on the updated ground truth indication of classification of maliciousness for the artifact, a representation of a performance of a source of classification from the plurality of sources of classification for that artifact.
11 . The one or more non-transitory processor readable media of claim 10 , wherein the artifact includes at least one of a file, a uniform resource locator (URL) or a device.
12 . The one or more non-transitory processor readable media of claim 10 , wherein the plurality of sources of classification includes at least one machine learning (ML) model trained to classify potentially malicious artifacts.
13 . The one or more non-transitory processor readable media of claim 10 , wherein the code to cause the one or more processors to determine includes code to cause the one or more processors to determine the ground truth indication of classification of maliciousness for the artifact based on a predetermined number of scores from the plurality of scores identifying the artifact as malicious.
14 . The one or more non-transitory processor readable media of claim 10 , wherein the at least one source of classification from the plurality of sources of classification includes a machine learning (ML) model trained to classify potentially malicious artifacts, the instructions further comprising code to cause the one or more processors to:
automatically retrain the ML model based on the updated ground truth indication of classification of maliciousness for the artifact.
15 . The one or more non-transitory processor readable media of claim 10 , wherein the code to cause the one or more processors to determine includes code to cause the one or more processors to determine the ground truth indication of classification of maliciousness for the artifact based on a number of sources of classification from the plurality of sources of classification associated with the artifact meeting a criterion.
16 . A method, comprising:
receiving information associated with a plurality of artifacts, the information associated with each artifact from the plurality of artifacts being based on a source of classification from a plurality of sources of classification;
selecting, based on a minimum number of sources of classification for each artifact from the plurality of artifacts, sample artifacts from the plurality of artifacts to define a sample set;
determining, based on the selecting, a ground truth indication of classification of maliciousness for each sample artifact in the sample set; and
displaying, via an interface and based on the ground truth indication of classification of maliciousness for one or more sample artifacts in the sample set, a representation of a performance of a source of classification from the plurality of sources of classification.
17 . The method of claim 16 , further comprising:
receiving, from a user and via the interface, an indication of the minimum number of sources of classification.
18 . The method of claim 16 , wherein each artifact from the plurality of artifacts includes at least one of a file, a uniform resource locator (URL) or a device.
19 . The method of claim 16 , wherein the plurality of sources of classification includes at least one machine learning (ML) model trained to classify potentially malicious artifacts.
20 . The method of claim 16 , further comprising:
receiving, via the interface, a threshold score associated with a classification of maliciousness for each artifact from the plurality of artifacts, the determining being based on the threshold score.