Saved in:
| Main Author: | |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2512.22740 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908801561526272 |
|---|---|
| author | Kang, Sungwoo |
| author_facet | Kang, Sungwoo |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_22740 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Multi-Task Learning for Metal Alloy Property Prediction: An Empirical Study of Negative Transfer and Mitigation Strategies Kang, Sungwoo Machine Learning 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. |
| title | Multi-Task Learning for Metal Alloy Property Prediction: An Empirical Study of Negative Transfer and Mitigation Strategies |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2512.22740 |