Pareto Optimization with Robust Evaluation for Noisy Subset Selection

A research team including Yiheng Xu and Chao Qian has published a new paper introducing PORE (Pareto Optimization with Robust Evaluation), a novel algorithm designed to solve the noisy subset selection problem. This is a fundamental challenge in combinatorial optimization where the goal is to select a limited-size subset from a larger set to maximize an objective function, but the function's evaluation is corrupted by real-world noise. Classic approaches like the greedy algorithm and even newer multi-objective evolutionary algorithms (MOEAs) like POSS and PONSS struggle with this noise or require excessive computational power. PORE directly addresses this by simultaneously maximizing a robust evaluation of the objective and minimizing the subset size, allowing it to efficiently identify high-quality, structured solutions.

Experiments on practical applications such as influence maximization in social networks and sparse regression for machine learning show that PORE significantly outperforms its predecessors. The algorithm's core innovation is its robust evaluation function, which was validated through ablation studies to be the key driver of its success. By providing a more reliable way to optimize under uncertainty, PORE offers a superior tool for data scientists and engineers working on problems where measurements or model evaluations are inherently imperfect, paving the way for more resilient AI and optimization systems.