Multi-Task Learning for Metal Alloy Property Prediction: An Empirical Study of Negative Transfer and Mitigation Strategies
This addresses the challenge of task-level imbalance in materials science for researchers, offering a strategic framework to optimize MTL for high-throughput screening, though it is incremental in refining existing techniques.
The study tackled the problem of negative transfer in multi-task learning for metal alloy property prediction, finding that MTL degraded regression performance for resistivity and hardness by up to 15% but improved classification recall for amorphous-forming ability by 10%. It traced this to mismatched functional forms causing gradient misalignment and proposed mitigation strategies like PCGrad to recover single-task performance.
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.