NeuroPareto: Calibrated Acquisition for Costly Many-Goal Search in Vast Parameter Spaces

TL;DR

NeuroPareto is a novel multi-objective optimization framework that efficiently balances exploration and exploitation in high-dimensional, costly search spaces using calibrated uncertainty estimation and hierarchical screening.

Contribution

It introduces a cohesive architecture combining rank-centric filtering, uncertainty disentanglement, and history-conditioned acquisition for improved Pareto optimization.

Findings

Outperforms baseline methods in Pareto proximity and hypervolume.

Effectively balances convergence and diversity in complex landscapes.

Maintains low computational overhead with hierarchical screening.

Abstract

The pursuit of optimal trade-offs in high-dimensional search spaces under stringent computational constraints poses a fundamental challenge for contemporary multi-objective optimization. We develop NeuroPareto, a cohesive architecture that integrates rank-centric filtering, uncertainty disentanglement, and history-conditioned acquisition strategies to navigate complex objective landscapes. A calibrated Bayesian classifier estimates epistemic uncertainty across non-domination tiers, enabling rapid generation of high-quality candidates with minimal evaluation cost. Deep Gaussian Process surrogates further separate predictive uncertainty into reducible and irreducible components, providing refined predictive means and risk-aware signals for downstream selection. A lightweight acquisition network, trained online from historical hypervolume improvements, guides expensive evaluations toward regions balancing convergence and diversity. With hierarchical screening and amortized surrogate updates, the method maintains accuracy while keeping computational overhead low. Experiments on DTLZ and ZDT suites and a subsurface energy extraction task show that NeuroPareto consistently outperforms classifier-enhanced and surrogate-assisted baselines in Pareto proximity and hypervolume.