Reproduce “LaMM: Semi-Supervised Pre-Training of Large-Scale Materials Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yosuke Oyama, Yusuke Majima, Eiji Ohta, Yasufumi Sakai
LaMM introduces a semi-supervised pre-training approach for Neural Network Potentials, leveraging approximately 300 million combined labeled and unlabeled material samples. This method reduces data preparation costs while enhancing model performance and training efficiency across diverse materials tasks.
Key insights
- Semi-supervised pre-training significantly improves the fine-tuning performance and accelerates convergence of Neural Network Potentials on downstream tasks.
- Effective load balancing in distributed training is crucial for heterogeneous materials datasets, leading to substantial speedups in NNP development.
- Pre-training on a large corpus of diverse materials enables models to learn fundamental chemical representations and periodic table relationships.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.