Multi-Task Learning for Metal Alloy Property Prediction: An Empirical Study of Negative Transfer and Mitigation Strategies

TL;DR

This study investigates multi-task learning for metal alloy property prediction, revealing that task mismatch causes negative transfer, and proposes strategies to mitigate this for different application needs.

Contribution

It demonstrates the limitations of MTL in materials science due to task mismatch and introduces mitigation techniques and a strategic framework for different use cases.

Findings

MTL degrades performance for resistivity and hardness

MTL improves recall for amorphous-forming ability

Gradient projection techniques recover single-task performance

Abstract

Multi-task learning (MTL) in materials science relies on the assumption that physically related properties share learnable representations. We challenge this assumption using a 54,028-sample metal alloy dataset exhibiting extreme task-level imbalance. Our results reveal a striking dichotomy: MTL significantly degrades regression performance for resistivity and hardness but improves classification recall for amorphous-forming ability. We trace this divergence to mismatched functional forms--such as resistivity's polynomial dependence versus hardness's complex interactions--which cause severe gradient misalignment during optimization. Evaluating Deep Imbalanced Regression techniques, we find that projecting conflicting gradients (PCGrad) recovers single-task performance, while combining label distribution smoothing with gradient normalization achieves the best overall balance. Consequently, we propose a strategic framework: utilize independent models for high-precision characterization, but employ MTL for high-throughput screening where recall is paramount. These findings support a "materials property clustering" hypothesis, suggesting that distinct physical mechanisms require specialized optimization strategies to overcome negative transfer.