MatFormBench: New Benchmark for Target-Driven Materials Formulation

Researchers have introduced MatFormBench, a benchmarking framework designed to evaluate generative strategies for target-driven materials formulation. Unlike existing machine learning benchmarks that focus solely on forward property prediction, MatFormBench addresses the critical gap in evaluating inverse optimization and generation algorithms. The framework integrates a physics-driven formulation generation scheme to create synthetic samples that emulate realistic materials structure-property relationships, with five escalating difficulty levels to quantify complexity. Additionally, MatFormScore, a multi-dimensional metric, comprehensively assesses algorithm performance. This work was published on arXiv (ID: 2605.26741) and aims to advance target-driven materials design by providing a systematic evaluation tool.