Protection Node for ML Classification Security

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Machine learning algorithms, particularly neural networks, are susceptible to malicious attacks where carefully crafted inputs can cause incorrect classifications, posing risks in applications like self-driving vehicles, and existing solutions like Generative Adversarial Networks require generating new datasets and do not account for previously successful malicious inputs.

Innovation Solution

A protection node that adds noise to incoming patterns, compares classifications from a clone machine learning algorithm and the original algorithm, and manages requests based on similarity rules to block or record malicious inputs, using these to improve the training of the original algorithm without the attacker's knowledge.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a machine learning algorithm processes original patterns directly, then classification accuracy is maintained, but the system becomes vulnerable to malicious requests causing incorrect classifications

Engineering Contradiction:

Improveclassification accuracyVSAvoidvulnerability to malicious requests

Core Design Contradiction:

ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent introduces a protection node as an intermediary between the user node and the machine learning algorithm. This protection node adds noise to incoming patterns and compares classifications from multiple clones before forwarding to the original algorithm, thereby mediating the interaction and protecting against malicious requests while maintaining classification accuracy

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The protection node performs preliminary actions by adding noise to patterns and obtaining classifications from clone algorithms before the original machine learning algorithm processes the request. This preliminary processing identifies potentially malicious inputs in advance, preventing them from reaching the original algorithm and causing incorrect classifications

Inventive Principle:

Principle #10Preliminary action

2Reliability

If noise is added to patterns and multiple classifications are compared, then protection against malicious requests is improved, but processing time and computational resources increase

Engineering Contradiction:

Improveprotection against malicious requestsVSAvoidprocessing time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The protection node performs partial classification checks by obtaining classifications from one or more clone algorithms without requiring all clones to fully process every request. The system uses a defined similarity rule that may allow some requests to pass through with minimal verification, reducing processing time while maintaining protection effectiveness

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The system changes parameters such as the amount of noise added to patterns and the threshold for classification similarity. By adjusting these parameters, the system can balance between thorough protection (higher reliability) and faster processing (lower time loss), optimizing performance based on specific operational requirements

Inventive Principle:

Principle #35Parameter changes

3Measurement precision

If multiple clones of the machine learning algorithm are used for comparison, then detection of malicious inputs is improved, but device complexity increases

Engineering Contradiction:

Improvedetection precision of malicious inputsVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The protection node creates simplified copies (clones) of the machine learning algorithm that can quickly process patterns and provide classifications for comparison. These clones are less complex than the original algorithm but sufficient for detecting malicious inputs through classification comparison, thereby improving detection precision without proportionally increasing overall system complexity

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The system segments the classification task by dividing it between multiple clone algorithms and the original machine learning algorithm. Each clone handles a portion of the analysis by providing a classification for comparison, while the original algorithm processes only those requests that pass the preliminary similarity check, reducing the complexity burden on any single component

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS11916931B2Methods for protecting pattern classification node from malicious requests and related networks and nodes

Publication Date: 2024.02.27 TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)

AI summary

A method of operating a protection node for protecting a pattern classification node from malicious requests may be provided. The protection node may receive, from a user node, a request containing an original pattern to be classified by a machine learning algorithm performed by the pattern classification node. The protection node may add noise to the original pattern to generate a noisy pattern. The protection node may obtain a first classification of the noisy pattern based on processing of the noisy pattern by a first clone of the machine learning algorithm at the protection node; obtain a second classification of the original pattern based forwarding the request for processing of the original pattern by the machine learning algorithm performed at the pattern classification node; and compare the first and second classifications to determine whether the first and second classifications satisfy a defined similarity rule. The protection node may use the comparison to manage the request from the user node.