Self-Organizing Map Valued Links for Trajectory Classification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing self-organizing maps struggle to efficiently classify and visualize the dynamic behavior of individuals in geographical and temporal contexts, particularly due to redundancy in location data and inefficiencies in tracing routes taken by mobile elements, such as users of communication networks or mobile devices with RF-ID tags, which leads to difficulties in unsupervised learning and automatic classification.

Innovation Solution

A self-organizing map with valued topological links that determines and exploits the valuation of interconnection links between neurons, allowing for unsupervised learning and automatic classification of spatio-temporal data, balancing positional and evolutionary influences, and reducing redundant trajectories by initializing synaptic weights and topological connections, updating prototype vectors and densities, and refining segmentation based on local densities and connectivity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional self-organizing maps are used to classify geographical and temporal data, then the system can process large datasets, but the classification efficiency is insufficient and redundant trajectories cannot be effectively reduced

Engineering Contradiction:
Improveclassification efficiencyVSAvoidredundant trajectories
Core Design Contradiction:
ProductivityVSLoss of information

Solution Approach 1:

The patent applies preliminary action by pre-initializing the synaptic weights of neurons using principal component analysis (PCA) before the actual self-organizing process. This preliminary step pre-processes the data structure and positions neurons in optimal initial configurations, enabling faster convergence and more efficient classification of spatio-temporal trajectories during the learning phase.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent changes key parameters of the self-organizing map by modifying the initialization method of synaptic weights from random to PCA-based deterministic values. This parameter change transforms the learning dynamics, allowing the network to quickly capture dominant variance directions in trajectory data and efficiently distinguish between redundant and meaningful movement patterns.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If conventional learning methods are applied to spatio-temporal data, then basic classification can be achieved, but the visualization of individual behaviors in cartographic representation is insufficient

Engineering Contradiction:
Improvebehavior classification accuracyVSAvoidcartographic visualization system
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies dimensionality change by mapping high-dimensional spatio-temporal trajectory data onto a two-dimensional cartographic representation through the self-organizing map's topological structure. This transformation preserves spatial relationships and temporal patterns while providing intuitive visual visualization of individual behaviors, enabling accurate classification without overwhelming system complexity.

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

Solution Approach 2:

The patent segments the continuous spatio-temporal data space into discrete topological neighborhoods and winning neuron regions. Each neuron represents a specific behavior pattern or location, and the segmentation allows independent analysis and visualization of different behavior types while maintaining their spatial and temporal relationships in the cartographic display.

Inventive Principle:
Principle #1Segmentation

3Productivity

If standard self-organizing maps process mobile element trajectories, then data can be collected, but the tracing of routes taken by mobile elements is inefficient

Engineering Contradiction:
Improvetrajectory processing speedVSAvoidroute tracing accuracy
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent implements feedback mechanisms where the self-organizing map continuously adjusts synaptic weights based on the difference between current neuron activations and target trajectory patterns. This iterative feedback process refines the route tracing accuracy over time while maintaining high processing speed through efficient vector calculations and topological updates.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent applies dynamics by making the synaptic weights and neighborhood structures adaptive and time-varying during the learning process. The system dynamically adjusts its internal parameters based on incoming trajectory data, enabling it to efficiently track and accurately trace evolving mobile element routes while maintaining computational performance.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP2240891B1Methods for updating and training for a self-organising card
Publication Date: 2013.02.13 PARIS 13 UNIVERSITY
  • EP2240891B1 patent drawingFigure 1~3
  • EP2240891B1 patent drawingFigure 4~5
  • EP2240891B1 patent drawingFigure 6~7

AI summary

The invention relates to an updating method that comprises selecting the best winning neurone and the second best winning neurone, modifying the prototype vectors of the best winning neurone and of the neurones located around the best winning neurone in the direction of the vector of the training point (x(k)), determining the neighbouring neurones (N(u*)) of the best winning neurone(u*), and, if the second best winning neurone (u**) is part of the neighbouring neurones (N(u*)), increasing the valuation of the connection between the first and second best winning neurones. The updating method further comprises reducing the valuation of each connection between the first best winning neurone and the directly neighbouring neurones (N(u*)) different from the second best winning neurone (u**).