Mixup-Augmented Meta-Learning for Sample-Efficient Fine-Tuning of Protein Simulators

• DOI: 10.48550/arxiv.2308.15116
• Publication Date: 2023-01-01
• Authors: Jingbang Chen; Yian Wang; Xingwei Qu; Shuangjia Zheng; Yaodong Yang; Hao Dong; Jie Fu
• Keywords: Initialization; Sample (material)

Molecular dynamics simulations have emerged as a fundamental instrument for studying biomolecules. At the same time, it is desirable to perform of collection particles under various conditions in which molecules can fluctuate. In this paper, we explore and adapt soft prompt-based learning method molecular tasks. Our model remarkably generalize unseen out-of-distribution scenarios with limited training data. While our work focuses on temperature test case, versatility approach allows efficient simulation through any continuous dynamic conditions, such pressure volumes. framework has two stages: 1) Pre-trains data mixing technique, augments structure prompts, then applies curriculum by increasing ratio them smoothly. 2) Meta-learning-based fine-tuning improves sample-efficiency process gives prompt-tuning better initialization points. Comprehensive experiments reveal that excels accuracy in-domain demonstrates strong generalization capabilities samples.