Distance-Based Replica Interaction Optimization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing optimization methods using multiple replicas in discrete optimization problems often fail to fully explore the state space due to replicas getting stuck in local solutions, especially when considering interactions between them.

Innovation Solution

An optimization device and method that selectively determines the interaction between replicas based on distance, using methods like periodic, group-based, dynamically adjusted, or random interaction selection to influence the proposal and acceptance probabilities in the Metropolis-Hastings method, thereby enhancing the exploration of the state space.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If replicas interact with all other replicas regardless of distance, then the exploration of state space is improved, but the computational burden increases

Engineering Contradiction:
Improvesolution search capabilityVSAvoidcomputational burden
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the replica interactions by introducing distance-based grouping, where replicas are divided into different groups based on their distance in the state space. Only replicas within certain distance thresholds interact with each other, rather than all replicas interacting universally. This segmentation reduces the number of interactions while maintaining solution search capability.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by making the interaction strength and probability dependent on the local distance between replicas in the state space. Replicas that are closer in the state space have higher interaction probability, while distant replicas have lower or zero interaction probability. This creates a non-uniform interaction pattern that reduces computational burden while preserving essential exploration.

Inventive Principle:
Principle #3Local quality

2Reliability

If replicas interact frequently with each other, then the solution search capability is improved, but replicas get stuck in local solutions

Engineering Contradiction:
Improvesolution search capabilityVSAvoidconvergence to local solution
Core Design Contradiction:
ReliabilityVSStability of the object's composition

Solution Approach 1:

The patent introduces dynamics by making the interaction probability between replicas dynamic rather than static. The interaction probability changes based on the current distance between replicas in the state space and follows a predetermined distribution pattern. This dynamic adjustment allows the system to adapt interaction strength during the optimization process, preventing premature convergence to local solutions while maintaining solution search capability.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent implements periodic action through the use of predetermined interaction probability distributions that create periodic patterns of interaction. Replicas interact with higher probability at certain distance intervals and lower probability at other intervals, creating a periodic interaction pattern that helps escape local solutions while maintaining efficient exploration.

Inventive Principle:
Principle #19Periodic action

3Device complexity

If distance-based interaction selection is implemented, then computational burden is reduced, but the exploration efficiency may be affected

Engineering Contradiction:
Improvecomputational burdenVSAvoidexploration efficiency
Core Design Contradiction:
Device complexityVSProductivity

Solution Approach 1:

The patent applies parameter changes by introducing distance as a key parameter that controls interaction behavior. The interaction probability is changed as a function of distance between replicas, using predetermined distributions. This parameter-based control allows the system to maintain exploration efficiency while reducing computational burden by focusing interactions on relevant replica pairs.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3975057A1Optimization device, optimization method, and optimization program
Publication Date: 2022.03.30 FUJITSU LTD
  • EP3975057A1 patent drawingFigure 1
  • EP3975057A1 patent drawingFigure 2
  • EP3975057A1 patent drawingFigure 3

AI summary

An optimization device performs: specifying, for each of a plurality of replicas each having a plurality of state variables, an amount of change in strength of interaction according to change in a distance between the replica and another replica, in a state space that indicates a space in which a combination of the values of the plurality of state variables may exist, in the case where a value of a first state variable among the plurality of state variables of the replica is updated; and determining, by using a proposal probability according to the amount of change in the strength of the interaction and an acceptance probability according to a target probability distribution in the case where the value of the first state variable is updated, whether or not to update the value of the first state variable.