Structured File Portion Impact Analysis for ML Malicious Content Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models struggle to provide insights into why they classify files as malicious, due to their complexity, making it difficult to understand the impact of specific file portions on classification outcomes.
Innovation Solution
A processor-based system that receives a structured file, removes portions to create modified files, extracts features, and inputs these to a machine learning model to identify the impact of each portion on classification, storing the results for analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are used for classifying malicious content, then classification capability is improved, but understanding of classification reasons deteriorates
Solution Approach 1:
The patent segments the file into multiple portions and analyzes each portion's impact on classification separately. This allows the complex model to be broken down into manageable components that can be individually evaluated, maintaining classification capability while improving interpretability of the classification reasons.
Solution Approach 2:
The patent introduces an intermediary analysis layer that examines the relationship between file portions and classification outcomes. This intermediary process generates explanations by analyzing how removing or modifying specific portions affects the model's classification, thereby bridging the gap between complex model operations and human-understandable reasons.
2Measurement precision
If machine learning models are used for classifying malicious content, then classification accuracy is improved, but interpretability of classification reasons deteriorates
Solution Approach 1:
The patent implements a feedback mechanism where the system analyzes the impact of removing or modifying file portions on classification outcomes. This feedback loop generates interpretable information about which file portions contribute most to the classification decision, thereby recovering interpretability information that would otherwise be lost in complex model operations.
Solution Approach 2:
The patent replaces the opaque mechanical decision-making process of the machine learning model with an analytical process that examines file portions systematically. This substitution transforms the black-box classification into an interpretable analysis that identifies specific file portions and their impact on the classification outcome.
3Loss of information
If portions of structured files are removed to analyze impact, then understanding of classification contributions is improved, but processing time increases
Solution Approach 1:
The patent applies partial action by selectively removing or modifying only specific portions of the file that are suspected to have significant impact on classification, rather than analyzing all possible modifications. This approach achieves sufficient classification insight while minimizing the time cost of extensive file manipulation and reprocessing.
Data Source
AI summary
In some embodiments, a non-transitory processor-readable medium stores code representing instructions to be executed by a processor. The code includes code to cause the processor to receive a structured file for which a machine learning model has made a malicious content classification. The code further includes code to remove a portion of the structured file to define a modified structured file that follows a format associated with a type of the structured file. The code further includes code to extract a set of features from the modified structured file. The code further includes code to provide the set of features as an input to the machine learning model to produce an output. The code further includes code to identify an impact of the portion of the structured file on the malicious content classification of the structured file based on the output.