Recurrent Neural Network Multi-Window Analysis for Robust Decisions

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Recurrent neural networks are susceptible to adversarial examples and noise due to their temporal dynamic behavior, leading to inconsistent and unreliable decision-making.

Innovation Solution

Implementing a method that defines multiple windows of time with varying durations and start times to identify and compare topological patterns of activity within a recurrent neural network, allowing for the classification and adjustment of less robust decisions, thereby enhancing the network's robustness and resilience to noise and adversarial attacks.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If recurrent neural networks process data using temporal dynamic behavior, then the network can capture sequential patterns and context over time, but the network becomes susceptible to adversarial examples and noise leading to inconsistent decision-making

Engineering Contradiction:
Improvetemporal dynamic behaviorVSAvoiddecision-making consistency
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent segments the temporal processing into multiple windows of time with different durations and start times. Each window captures topological patterns independently, allowing the system to analyze temporal dynamics at different scales while reducing susceptibility to adversarial examples that target specific temporal configurations.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent uses multiple overlapping windows that extend beyond the minimum necessary temporal coverage. This excessive sampling of temporal patterns ensures that relevant information is captured across different time scales, while irrelevant or adversarial inputs are attenuated through the comparison process across windows.

Inventive Principle:
Principle #16Partial or excessive action

2Productivity

If the network uses fixed time windows for processing, then the processing is simple and fast, but the network cannot capture patterns at different temporal scales or respond to varying input dynamics

Engineering Contradiction:
Improveprocessing speedVSAvoidtemporal scale flexibility
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent implements dynamic time windows with varying durations and start times that adapt to the temporal characteristics of the input data. This allows the network to capture patterns at multiple temporal scales while maintaining efficient processing through the structured comparison framework.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent adds the dimension of temporal scale by using multiple windows with different durations. This transforms the single-time-scale processing into multi-scale temporal analysis, enabling the network to capture both short-term and long-term patterns simultaneously.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

3Reliability

If the network reinforces all processing results equally, then the decision-making is consistent, but relevant information cannot be distinguished from irrelevant or adversarial inputs

Engineering Contradiction:
Improvedecision consistencyVSAvoidsignal-to-noise discrimination
Core Design Contradiction:
ReliabilityVSLoss of information

Solution Approach 1:

The patent implements feedback through the comparison of topological patterns across multiple time windows. Processing results that are consistent across windows are reinforced, while those that vary are attenuated. This feedback mechanism enables the network to distinguish relevant information from noise and adversarial inputs while maintaining decision consistency.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent changes the parameter of temporal scale by analyzing patterns at multiple time window durations. This parameter variation allows the network to identify which processing results are robust across different temporal scales (relevant information) versus those that are scale-dependent (noise or adversarial inputs).

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4579525A1Interpreting and improving the processing results of recurrent neural networks
Publication Date: 2025.07.02 INAIT SA
  • EP4579525A1 patent drawingFigure 1
  • EP4579525A1 patent drawingFigure 2~3
  • EP4579525A1 patent drawingFigure 4

AI summary

A method includes defining a plurality of different windows of time in a recurrent artificial neural network, wherein each of the different windows has different durations, has different start times, or has both different durations and different start times, identifying occurrences of topological patterns of activity in the recurrent artificial neural network in the different windows of time, comparing the occurrences of the topological patterns of activity in the different windows, and classifying, based on a result of the comparison, a first decision that is represented by a first topological pattern of activity that occurs in a first of the windows as less robust than a second decision that is represented by a second topological pattern of activity that occurs in a second of the windows.