New FMQA framework uses stage-dependent encoding for black-box optimization

Researchers have developed a stage-dependent framework for factorization machine with quadratic-optimization annealing (FMQA) to improve black-box optimization. This new approach allows for different integer-binary encodings to be used during the surrogate learning phase and the Ising machine solution search phase, addressing the limitation of conventional FMQA which uses a single encoding throughout. The proposed framework, evaluated on the Rastrigin function, demonstrated that one-hot encoding is crucial for learning stage performance, consistently yielding lower residual errors. Further improvements were observed by switching to domain-wall encoding for the solution search phase under finer discretization levels. AI

IMPACT This research could lead to more efficient optimization techniques for complex problems in machine learning.

RANK_REASON The item is a research paper detailing a new optimization framework. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• binary code
• domain-wall encoding
• Ising Machines for Diophantine Problems in Physics
• OhDw
• one-hot encoding
• Rastrigin function

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →