ML-Guided Improvement Search

Neural-Guided Improvement Search (NGIS): Enhancing Optimization with AI

Improvement Scheme

Neural-Guided Improvement Search (NGIS) improves solutions by combining traditional search methods with deep neural networks. It uses individual local search techniques to incrementally refine candidate solutions, enhanced by GPU-based parallelization for faster computation. Population-based search methods, like evolutionary algorithms, help explore broader solution spaces by evolving and recombining candidates. Memetic algorithms, blending population-based search with local refinement, further boost the ability to reach near-optimal solutions.

Learning Scheme

NGIS learning is powered by meta-learning and reinforcement learning. Meta-learning improves generalization across problem types, preventing overfitting. Reinforcement learning, specifically off-policy methods, trains neural networks to guide the improvement process. By learning from the outcomes of refined solutions, NGIS dynamically adjusts its strategies, leading to more efficient and effective search processes.