Reinforcement Learning Graph Segmentation for Combinatorial Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional methods for solving combinatorial optimization problems, such as the Traveling Salesman Problem, require significant time and computing resources, making them inefficient for complex optimization tasks in operational research.
Innovation Solution
A method involving a reinforcement learning model applied to an initial graph, where edges are selected and deleted to generate a partial solution, followed by an annealer-based solver to reconnect disjoint segments and generate a new solution, with the model being re-trained for improved solutions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If traditional optimization methods are used to solve combinatorial optimization problems, then solution accuracy is maintained, but computational time and resource consumption increase significantly
Solution Approach 1:
The patent divides the combinatorial optimization problem into multiple sub-problems or sub-tours. The annealer-based solver processes these segmented sub-problems independently and combines them to form the complete solution. This segmentation reduces the computational complexity and time required while maintaining solution quality through systematic recombination of sub-solutions.
Solution Approach 2:
The reinforcement learning model performs preliminary actions by pre-processing the initial solution, identifying and selecting specific edges or segments before the annealer-based solver processes them. This preliminary filtering and preparation reduces the search space for the optimization solver, thereby decreasing computational time while preserving the essential structure needed for solution quality.
2Productivity
If traditional optimization methods are applied to complex processes with multiple constraints, then comprehensive constraint satisfaction is achieved, but computing resources are consumed excessively
Solution Approach 1:
The patent segments the complex optimization problem into manageable sub-problems that can be processed independently by the annealer-based solver. This division reduces the computational burden on any single processing unit, lowering energy consumption while maintaining overall optimization efficiency through parallel or sequential processing of segments.
Solution Approach 2:
The reinforcement learning model acts as an intermediary between the initial solution and the annealer-based solver. It pre-processes the input by selecting and filtering edges based on learned patterns, thereby reducing the complexity of the problem presented to the energy-intensive annealer solver. This intermediary step improves productivity by preparing data in a form that requires fewer computing resources for optimization.
3Measurement precision
If the reinforcement learning model processes the entire graph directly, then complete solution coverage is achieved, but computational complexity increases
Solution Approach 1:
The reinforcement learning model extracts and selects only the most relevant edges from the complete graph based on the initial solution and learned patterns. By taking out and focusing on specific critical edges rather than processing the entire graph, the computational complexity is reduced while the solution remains complete because the selected edges contain the essential information needed for optimal routing.
Solution Approach 2:
The patent applies partial action by having the reinforcement learning model process only a subset of edges selected from the complete graph, rather than analyzing every edge. This partial processing reduces computational complexity while still achieving complete solution coverage because the selected edges are sufficient to construct the optimal tour when combined with the annealer-based solver's optimization capabilities.
Data Source
AI summary
In an embodiment, a first graph corresponding to an initial solution of a combinatorial optimization problem is received. A reinforcement learning (RL) model is applied on the received first graph. A predefined number of a set of edges is selected from the received first graph. The selected set of edges is deleted from the received first graph to generate a second graph, based on a disconnection of a set of segments associated with the selected set of edges. The generated second graph corresponds to a partial solution. Thereafter, a partial tour may be determined using an annealer-based solver to generate a third graph, based on a connection of the predefined number of a set of disjoint segments. The generated third graph corresponds to a new solution. The RL model is re-trained to determine an improved solution. The determined improved solution is rendered on a display device.