Attention-Based Malware Detection From API Behavior Sequences

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional malware detection methods struggle to reliably identify malicious software due to camouflage and the rapid growth of applications, making it difficult for end users to detect and classify malware without user interaction, and existing systems often require execution in unsecured environments, risking contamination.

Innovation Solution

A method using a neural network-based classifier with an attention mechanism to analyze behavior sequences, such as API-calls, to identify malware, focusing on significant subsequences for classification, and incorporating expert feedback for supervised learning.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional malware detection methods are used, then the detection process is simple, but the detection accuracy is low due to malware camouflage and rapid application growth

Engineering Contradiction:
Improvemalware detection accuracyVSAvoiddetection system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The behavior sequence is divided into multiple subsequences, and the classification is performed on these segmented parts rather than the entire sequence at once. This segmentation allows the system to focus on relevant portions of the behavior sequence that are indicative of malware, thereby improving detection accuracy while managing complexity through localized analysis.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

An attention mechanism is introduced as an intermediary component between the behavior sequence and the classifier. This attention mechanism automatically identifies and weights the most relevant subsequences, acting as a mediator that filters and prioritizes information before it reaches the classification stage, thus improving accuracy without requiring manual complexity management.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If the entire behavior sequence is analyzed for classification, then comprehensive detection is achieved, but the data processing load and time consumption increase significantly

Engineering Contradiction:
Improveclassification reliabilityVSAvoidanalysis time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

Instead of analyzing the entire behavior sequence, the system applies partial action by focusing only on the most relevant subsequences identified through the attention mechanism. This partial analysis maintains classification reliability by concentrating computational resources on the most informative parts of the sequence, thereby reducing overall analysis time without sacrificing detection accuracy.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The attention mechanism extracts and isolates the most relevant subsequences from the full behavior sequence, separating the critical information from the less important parts. This extraction process allows the classifier to work with a reduced, high-value subset of data, maintaining reliability while significantly reducing processing time and computational load.

Inventive Principle:
Principle #2Taking out (Extraction)

3Measurement precision

If expert verification of classification results is implemented, then detection accuracy improves, but the process requires more user interaction and time

Engineering Contradiction:
Improveclassification accuracyVSAvoidoperational simplicity
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The system partially automates the expert verification process by using the attention mechanism to pre-identify and present only the most relevant subsequences to experts for verification. This self-service approach filters out less important information before it reaches the expert, reducing the time and effort required for verification while maintaining improved accuracy through targeted expert review.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The attention mechanism performs preliminary action by pre-processing and prioritizing the behavior sequence before it reaches the expert verification stage. It automatically identifies and highlights the most relevant subsequences, so when experts do verify the results, they are already focused on the critical portions, thereby improving accuracy without proportionally increasing the time and interaction required.

Inventive Principle:
Principle #10Preliminary action

4Measurement precision

If neural network-based classification is used, then detection accuracy improves, but the computational requirements and system complexity increase

Engineering Contradiction:
Improvemalware classification accuracyVSAvoidcomputational energy consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The neural network processes segmented subsequences rather than the entire behavior sequence at once. This segmentation reduces the computational burden on the neural network at each processing step, allowing high accuracy to be achieved through multiple focused processing passes on smaller data chunks rather than a single pass on the complete sequence, thereby managing energy consumption more efficiently.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The attention mechanism serves as an intermediary that pre-filters and weights the input data before it reaches the neural network. By providing the neural network with already-prioritized, relevant subsequences, the attention mechanism reduces the computational energy required for the neural network to process irrelevant or less important information, while still achieving high classification accuracy on the critical data.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3970043B1Apparatuses and methods for detecting malware
Publication Date: 2025.10.01 HUAWEI TECH CO LTD
  • EP3970043B1 patent drawingFigure 1
  • EP3970043B1 patent drawingFigure 2
  • EP3970043B1 patent drawingFigure 3

AI summary

Apparatuses and methods for determining if a computer program is malware and to which malware class it belongs to. In the method the behaviour of a computer program is traced by observing the activity of the program. Behaviour sequences comprising API-calls or similar activity of a computer program are then provided into a classifier for classifying the computer program. From the outcome of the classifier a classification result and the portions relevant to decision can be provided to a person for further confirmation.