Quantum Annealing Time to Solution Prediction via Piecewise Curves
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Quantum annealing systems face challenges in efficiently estimating or predicting the time to solution for optimization problems, particularly due to the rarity of optimal solutions for large problem instances and the computational burden of proving optimality.
Innovation Solution
The development of a system that includes an orchestration engine and a prediction engine to estimate or predict the time to solution or time to best known solution in quantum annealing systems. This system collects data from multiple annealing executions with varying hyperparameters, trains prediction engines to generate piecewise functions representing the time to solution curve, and selects the most appropriate quantum annealing system based on predicted times.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optimal solutions are used as target solutions for time to solution experiments, then the accuracy of time to solution measurement is improved, but the computational burden and availability of target solutions deteriorates
Solution Approach 1:
The patent replaces the need for expensive and rare optimal solutions with readily available approximate solutions. By using approximate solutions as targets, the system can conduct multiple time to solution experiments without the computational burden of verifying optimality, thus resolving the contradiction between measurement accuracy and computational complexity
Solution Approach 2:
The patent creates a surrogate measurement system by copying the time to solution concept from optimal solution-based experiments to approximate solution-based experiments. This allows the system to measure time to solution using easily obtainable approximate solutions instead of difficult-to-verify optimal solutions, maintaining measurement validity while reducing computational burden
2Reliability
If multiple runs are performed to analyze search effort and identify solutions, then the reliability of solution identification is improved, but the time required for analysis deteriorates
Solution Approach 1:
The patent performs preliminary measurements of time to solution using approximate solutions before final solution identification. By conducting preliminary experiments with easier-to-obtain approximate solutions, the system can gather statistical data about search effort and solution identification time without requiring multiple complete runs to optimal solutions, thus improving reliability while reducing total analysis time
3Measurement precision
If time to solution is measured using optimal solutions, then the accuracy of performance measurement is improved, but the availability of target solutions deteriorates
Solution Approach 1:
The patent substitutes rare and unavailable optimal solutions with abundant approximate solutions that can be easily obtained during quantum annealing runs. This replacement maintains the ability to conduct performance measurements across different problem instances and quantum annealers while eliminating the availability constraint imposed by optimal solutions
Solution Approach 2:
The patent changes the target solution parameter from optimal to approximate solutions. This parameter change allows the time to solution measurement to be conducted using solutions that are both available and sufficient for performance evaluation, resolving the contradiction between measurement accuracy and solution availability
Data Source
AI summary
Systems and methods for predicting or estimating time to solution in quantum computing system. A prediction engine is trained to estimate a time to best known solution curve associated with a quantum annealing system. Using the estimated or predicted curve, when a new QUBO or Ising model is to be executed, the time to best known solution can be predicted or estimated. An orchestration engine may select the quantum annealing system that has the best time to best known solution.