Linear Model Partitioning for Faster Large-Scale Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer systems and networks face challenges in managing and processing large volumes of data related to electronic transactions, particularly in electronic trading systems, due to limited computing resources and high communication bandwidth requirements.
Innovation Solution
A computer-implemented system and process that simplifies and improves the computational performance by applying linear optimization algorithms to complex electronic multivariate data processing systems, using iterative partitioning of large models into sub-models, which are solved independently and relatively quickly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If linear optimization algorithms are applied to solve the full model directly, then solution accuracy is improved, but computational time and resource consumption increase significantly
Solution Approach 1:
The patent divides the full linear optimization model into multiple sub-models that can be solved independently and in parallel. This segmentation allows the system to maintain solution accuracy while significantly reducing computational time by distributing the solving process across multiple processing units.
Solution Approach 2:
The patent solves multiple sub-models partially or independently rather than solving the complete full model to optimality. By finding satisfactory solutions to sub-models and combining them, the system achieves adequate overall solution accuracy with reduced computational effort.
2Productivity
If more computing resources are allocated to process electronic transaction data, then data processing capability is improved, but hardware costs and system complexity increase
Solution Approach 1:
The patent segments the large-scale data processing task into smaller sub-models that can be processed independently. This approach enables the system to handle large volumes of electronic transaction data using distributed computing resources, reducing the complexity burden on any single computing node.
Solution Approach 2:
The patent creates a framework that can process different types of electronic transaction data using the same sub-modeling approach. This universal method allows the system to handle various data formats and transaction types without requiring specialized hardware for each data type.
3Reliability
If the full model is solved to obtain an optimized solution, then solution quality is improved, but computational resource consumption increases
Solution Approach 1:
The patent divides the full model into sub-models that consume fewer computational resources to solve. By distributing the computational load across multiple smaller problems, the system maintains solution quality while reducing the energy and resource consumption required for each individual solving operation.
Solution Approach 2:
The patent accepts that solving sub-models may not achieve the absolute optimal solution of the full model, but the partial solutions combined from sub-models provide sufficiently high solution quality for practical applications, thereby reducing overall resource consumption.
Data Source
AI summary
The disclosed embodiments related to multilateral portfolio compression using general large-scale linear optimization which pre-processes a model to decrease model size using domain knowledge to remove variables to reduce dimensionality, thereby making the model faster to solve and improving numerical characteristics. but it would not remove, for example, as much as half of the model, but rather a smaller fraction. The disclosed pre-processing enables an approximate solution for large, linear optimization models by automatically iteratively and selectively partitioning them into independently easily solvable sub-models. The sub-models are themselves linear optimization models, which can be solved with any preferred algorithm or library. The solutions for each sub-model are aggregated to obtain an acceptable, e.g., approximate, solution for a large model without solving the full model. At each iteration the disclosed embodiments will have a valid, feasible solution, if the user is satisfied before full convergence.