System and method for generating a convolution function for training a malware detection model
Disclosed are systems and methods generating a convolution function for training a malware detection model. An example method comprises selecting, by a processor, one or more commands from a log according to a set of predetermined rules, forming, by the processor, one or more behavior patterns from the one or more selected commands, determining, by the processor, a feature vector according to the one or more behavior patterns, generating, by the processor, a convolution function according to the feature vector, wherein a size of a result of the convolution function of the feature vector is less than the size of the feature vector, and computing, by the processor, one or more parameters for training a malware detection model using the convolution function on the one or more behavior patterns.
1. A method for generating a hash function for training a malware detection model, comprising:
selecting, by a processor, one or more commands from a behavior log according to a set of predetermined rules;
forming, by the processor, one or more behavior patterns from the one or more selected commands, wherein each behavior pattern comprises the one or more commands selected from the behavior log and one or more parameters describing the one or more selected commands;
determining, by the processor, a feature vector of the one or more behavior patterns;
generating, by the processor, a hash function of the feature vector using an auto-encoder, wherein a size of a result of the hash function is less than the size of the feature vector, and wherein a degree of similarity of the feature vector and a result of an inverse of the hash function of the result of the hash function of the feature vector is greater than a predetermined value; and
computing, by the processor, one or more parameters for training a malware detection model using the hash function on the one or more behavior patterns.
2. The method of claim 1 , wherein the feature vector is a bloom filter.
3. The method of claim 1 , further comprising:
forming the hash function using a metric learning method.
4. The method of claim 1 , wherein the set of predetermined rules comprise one or more of: 1) selecting the commands and/or parameters after a predetermined number of entries in the behavior log, 2) selecting the commands and/or parameters after a predetermined period of time from a previous command in the behavior log, 3) selecting the commands and/or parameters after a predetermined period of time from execution of a file under analysis, that are in a predetermined list, 4) selecting a first parameter of commands that take parameters greater than a predetermined threshold value, and 5) selecting a predetermined number of random parameters of commands that take parameters greater than a predetermined threshold value.
5. The method of claim 1 ,
wherein an output of the auto-encoder comprises data having a coefficient of similarity to the input data greater than a predetermined threshold value.
6. A system for generating a hash function for training a malware detection model, comprising, comprising:
a hardware processor configured to:
select one or more commands from a behavior log according to a set of predetermined rules;
form one or more behavior patterns from the one or more selected commands wherein each behavior pattern comprises the one or more commands selected from the behavior log and one or more parameters describing the one or more selected commands;
determine a feature vector of the one or more behavior patterns;
generate a hash function of the feature vector using an auto-encoder, wherein a size of a result of the hash function is less than the size of the feature vector, and wherein a degree of similarity of the feature vector and the result of an inverse of the hash function of the result of the hash function of the feature vector is greater than a predetermined value; and
compute one or more parameters for training a malware detection model using the hash function on the one or more behavior patterns.
7. The system of claim 6 , wherein the feature vector is a bloom filter.
8. The system of claim 6 , the hardware processor further configured to:
form the hash function using a metric learning method.
9. The system of claim 6 , wherein the set of predetermined rules comprise one or more of: 1) selecting the commands and/or parameters after a predetermined number of entries in the behavior log, 2) selecting the commands and/or parameters after a predetermined period of time from a previous command in the behavior log, 3) selecting the commands and/or parameters after a predetermined period of time from execution of a file under analysis, that are in a predetermined list, 4) selecting a first parameter of commands that take parameters greater than a predetermined threshold value, and 5) selecting a predetermined number of random parameters of commands that take parameters greater than a predetermined threshold value.
10. The system of claim 6 ,
wherein an output of the auto-encoder comprises data having a coefficient of similarity to the input data greater than a predetermined threshold value.
11. A non-transitory computer-readable medium, storing instructions thereon, which when executed perform a method for generating a hash function for training a malware detection model, comprising:
selecting, by a processor, one or more commands from a behavior log according to a set of predetermined rules;
forming, by the processor, one or more behavior patterns from the one or more selected commands wherein each behavior pattern comprises the one or more commands selected from the behavior log and one or more parameters describing the one or more selected commands;
determining, by the processor, a feature vector of the one or more behavior patterns;
generating, by the processor, a hash function of the feature vector using an auto-encoder, wherein a size of a result of the hash function is less than the size of the feature vector, and wherein a degree of similarity of the feature vector and the result of an inverse of the hash function of the result of the hash function of the feature vector is greater than a predetermined value; and
computing, by the processor, one or more parameters for training a malware detection model using the hash function on the one or more behavior patterns.
12. The medium of claim 11 , wherein the feature vector is a bloom filter.
13. The medium of claim 11 , further comprising:
forming the hash function using a metric learning method.
14. The medium of claim 11 , wherein the set of predetermined rules comprise one or more of: 1) selecting the commands and/or parameters after a predetermined number of entries in the behavior log, 2) selecting the commands and/or parameters after a predetermined period of time from a previous command in the behavior log, 3) selecting the commands and/or parameters after a predetermined period of time from execution of a file under analysis, that are in a predetermined list, 4) selecting a first parameter of commands that take parameters greater than a predetermined threshold value, and 5) selecting a predetermined number of random parameters of commands that take parameters greater than a predetermined threshold value.
15. The medium of claim 11 ,
wherein an output of the auto-encoder comprises data having a coefficient of similarity to the input data greater than a predetermined threshold value.