Ising Model Coefficient Compression for Memory-Constrained Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current optimization apparatuses face memory limitations when handling large-scale multivariable optimization problems due to the high number of coupling coefficients required for Ising models, which exceed the storage capacity of on-chip memories.
Innovation Solution
The apparatus compresses coupling coefficients based on symmetry or pattern properties of the coefficient matrix, reducing storage needs by storing only the diagonal and symmetric components, and decoding these components as needed for simulated annealing operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If coupling coefficients are stored in internal memory of the optimization apparatus, then access speed is improved, but storage capacity is insufficient for large-scale problems
Solution Approach 1:
The coefficient matrix is segmented into symmetric components and diagonal components, allowing selective storage of only the essential parts. The symmetric components are stored in a compressed format that divides the full matrix into manageable segments, fitting within internal memory capacity while maintaining fast access to all necessary coupling coefficients through computational reconstruction.
Solution Approach 2:
The invention extracts and stores only the diagonal components and symmetric components of the coefficient matrix separately from the full matrix. By taking out and storing only these essential components in a compressed format, the system reduces storage requirements to fit within internal memory while still enabling complete coefficient retrieval for optimization calculations.
2Measurement precision
If the number of bits for representing coupling coefficients is increased, then precision is improved, but memory storage requirements increase
Solution Approach 1:
The system stores coupling coefficients with sufficient precision for the optimization task at hand, rather than using maximum possible precision for all coefficients. By applying partial precision where needed and leveraging the compressed storage of symmetric components, the system achieves adequate measurement precision while keeping memory storage requirements within acceptable limits for internal memory.
Data Source
AI summary
An optimization apparatus includes a compression unit, a storage unit, a decoding unit, and an annealing unit. The compression unit outputs compressed coefficient data in which a coefficient matrix including coupling coefficients, each of which indicates a coupling strength between bits of an Ising model obtained by converting a calculation target problem, is compressed on the basis of a symmetry property or a pattern property of the coefficient matrix. The storage unit holds the compressed coefficient data outputted by the compression unit. The decoding unit decodes the compressed coefficient data stored in the storage unit to obtain the coupling coefficients. The annealing unit performs a simulated annealing operation by using the coupling coefficients obtained by the decoding unit.