Anomaly Detector ML Classifier Attack Protection via Manifold Subspace Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Anomaly detectors in cybersecurity are vulnerable to attacks where attackers manipulate input data points to appear normal when they are actually anomalous, leading to undetected cyberattacks and damage to computer systems.
Innovation Solution
The method involves identifying training data points in a manifold space, dividing it into subspaces, training subclassifiers for each subspace to determine decision boundaries, and using these subclassifiers to identify and protect against attacks by classifying input data points as anomalous or normal, with remedial actions such as blocking network access or rolling back changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single anomaly detector machine learning classifier is used to classify data points, then the device complexity is low, but the measurement precision of detecting attacks decreases
Solution Approach 1:
The patent divides the manifold space into multiple subspaces and trains a separate subclassifier for each subspace. This segmentation allows each subclassifier to specialize in detecting anomalies within its specific subspace, improving overall attack detection accuracy while maintaining manageable complexity through distributed classification
2Measurement precision
If multiple subclassifiers are trained for different subspaces to improve detection accuracy, then the measurement precision increases, but the device complexity increases
Solution Approach 1:
The patent transforms the classification problem from a single high-dimensional space into multiple lower-dimensional subspaces. By projecting data points onto different subspaces and training specialized subclassifiers for each, the system achieves better detection accuracy while reducing the effective complexity each classifier must handle
3Reliability
If the anomaly detector uses a simple classification boundary, then the device complexity is low, but the reliability against attacks decreases
Solution Approach 1:
The patent applies different decision boundaries to different subspaces rather than using a single global boundary. Each subclassifier learns local characteristics and decision boundaries specific to its subspace, improving robustness against attacks that exploit global boundary weaknesses while keeping individual local boundaries relatively simple
Data Source
AI summary
Identifying and protecting against an attack against an anomaly detector machine learning classifier (ADMLC). In some embodiments, a method may include identifying training data points in a manifold space for an ADMLC, dividing the manifold space into multiple subspaces, merging each of the training data points into one of the multiple subspaces, training a subclassifier for each of the multiple subspaces to determine a decision boundary for each of the multiple subspaces between normal training data points and anomalous training data points, receiving an input data point into the ADMLC, determining whether the input data point is an attack on the ADMLC due to a threshold number of the subclassifiers classifying the input data point as an anomalous input data point, and, in response to identifying the attack against the ADMLC, protecting against the attack.