# US 2021/0264024 A1 · Projected Vector Overflow Penalty as Mitigation for Machine Learning Model String Stuffing

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0264024 A1
- Document ID: US20210264024A1
- Status: Published application
- Kind: A1
- Application number: 16/798,120
- Filed: 2020-02-21
- Published: 2021-08-26
- Earliest claimed: 2020-02-21
- Assignees: Cylance Inc.
- Applicants: Cylance Inc.
- Inventors: Eric Glen Petersen
- Web page: https://patentgrep.com/patent/US20210264024A1

## CPC

- G06F 21/554: {involving event detection and direct action}
- G06F 21/562: {Static detection}
- G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
- G06N 5/04: Inference or reasoning models
- G06N 20/00: Machine learning
- G06F 2221/033: Test or assess software
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

An artifact is received from which features are extracted and used to populate a vector. The features in the vector are then reduced using a feature reduction operation to result in a modified vector having a plurality of buckets. Features within the buckets of the modified vector above a pre-determined projected bucket clipping threshold are then identified. Using the identified features, and overflow vector is then generated. The modified vector is then input into a classification model to generate a score. This score is adjusted based on the overflow vector and can then be provided to a consuming application or process. Related apparatus, systems, techniques and articles are also described.

## Claims

### Claim 1

A computer-implemented method comprising:
receiving an artifact;
extracting features from the artifact and populating a vector;
reducing features in the vector using a feature reduction operation to result in a modified vector having a plurality of buckets;
identifying features within buckets of the modified vector above a pre-determined projected bucket clipping threshold;
generating an overflow vector based on the identified features;
inputting the modified vector into a classification model to generate a score;
adjusting the score based on the overflow vector; and
providing the adjusted score to a consuming application or process.

### Claim 2 (depends on 1)

The method of claim 1, wherein the classification model characterizes the artifact as being malicious or benign to access, execute, or continue to execute.

### Claim 3 (depends on 2)

The method of claim 2 further comprising: preventing access or execution of the artifact when the classification model characterizes the artifact as being malicious.

### Claim 4 (depends on 1)

The method of claim 1, wherein the classification model is a machine learning model trained using a training data set and providing a continuous scale output.

### Claim 5 (depends on 1)

The method of claim 1, wherein the machine learning model comprises one or more of: a logistic regression model, a neural network, a convolutional neural network, a recurrent neural network, a generative adversarial network, a support vector machine, a random forest, or a Bayesian model.

### Claim 6 (depends on 1)

The method of claim 1, wherein the features comprises alphanumeric strings.

### Claim 7 (depends on 1)

The method of claim 1 further comprising:
inputting a plurality of vectorized malware samples into the classification model;
obtaining a plurality of scores based on the inputted vectorized malware samples; and
identifying, based on the classifications, buckets of the vectorized malware samples that influence the scores above the pre-determined threshold.

### Claim 8 (depends on 1)

The method of claim 1, wherein the classification model is a machine learning-based penalty model trained using training data that is synthesized by stuffing strings into benign and malware samples and providing a continuous scale output.

### Claim 9 (depends on 1)

The method of claim 1 further comprising:
calculating, using the overflow vector, a broad overflow summation totaling a number of buckets having features exceeding the pre-determined projected bucket clipping threshold.

### Claim 10 (depends on 9)

The method of claim 9 further comprising:
calculating, using the overflow vector, a weighted overflow summation in which certain buckets are weighted based on an empirical determination of how such buckets influence the score.

### Claim 11 (depends on 10)

The method of claim 10, wherein the score is adjusted based on a heuristic applying the broad overflow summation and the weighted overflow summation.

### Claim 12 (depends on 10)

The method of claim 10, wherein the score is adjusted by inputting both of the broad overflow summation and the weighted overflow summation into a machine learning model.

### Claim 13 (depends on 1)

The method of claim 1 further comprising:
clipping features within the modified vector prior to generating the overflow vector.

### Claim 14 (depends on 1)

The method of claim 1, wherein the feature reduction operation comprises random projection matrices.

### Claim 15 (depends on 1)

The method of claim 1, wherein the feature reduction operation comprises principal component analysis.

### Claim 16

A system comprising:
at least one data processor; and
memory storing instructions which, when executed by the at least one data processor, result in operations comprising:
receiving an artifact;
extracting features from the artifact and populating a vector;
reducing features in the vector using a feature reduction operation to result in a modified vector having a plurality of buckets;
identifying features within buckets of the modified vector above a pre-determined projected bucket clipping threshold;
generating an overflow vector based on the identified features;
inputting the modified vector into a classification model to generate a score;
adjusting the score based on the overflow vector; and
providing the adjusted score to a consuming application or process.

### Claim 17 (depends on 16)

The system of claim 16, wherein the classification model characterizes the artifact as being malicious or benign to access, execute, or continue to execute.

### Claim 18 (depends on 17)

The system of claim 17, wherein the operations further comprise:
preventing access or execution of the artifact when the classification model characterizes the artifact as being malicious.

### Claim 19

A non-transitory computer program product storing instructions which, when executed by at least one computing device, result in operations comprising:
receiving an artifact;
extracting features from the artifact and populating a vector;
reducing features in the vector using a feature reduction operation to result in a modified vector having a plurality of buckets;
identifying features within buckets of the modified vector above a pre-determined projected bucket clipping threshold;
generating an overflow vector based on the identified features;
inputting the modified vector into a classification model to generate a score;
adjusting the score based on the overflow vector; and
providing the adjusted score to a consuming application or process.

### Claim 20 (depends on 19)

The non-transitory computer program product of claim 19, wherein the classification model characterizes the artifact as being malicious or benign to access, execute, or continue to execute and the operations further comprise:
preventing access or execution of the artifact when the classification model characterizes the artifact as being malicious.
