Condensed Matter > Materials Science
[Submitted on 17 Jan 2024]
Title:Material Informatics through Neural Networks on Ab-Initio Electron Charge Densities: the Role of Transfer Learning
View PDF HTML (experimental)Abstract:In this work, the dynamic realms of Materials Science and Computer Science advancements meet the critical challenge of identifying efficient descriptors capable of capturing the essential features of physical systems. Such task has remained formidable, with solutions often involving ad-hoc scalar and vectorial sets of materials properties, making optimization and transferability challenging. We extract representations directly from ab-initio differential electron charge density profiles using Neural Networks, highlighting the pivotal role of transfer learning in such task. Firstly, we demonstrate significant improvements in regression of a specific defected-materials property with respect to training a deep network from scratch, both in terms of predictions and their reproducibilities, by considering various pre-trained models and selecting the optimal one after fine-tuning. The remarkable performances obtained confirmed the transferability of the existent pre-trained Convolutional Neural Networks (CNNs) on physics domain data, very different from the original training data. Secondly, we demonstrate a saturation in the regression capabilities of computer vision models towards properties of an extensive variety of undefected systems, and how it can be overcome with the help of large language model (LLM) transformers, with as little text information as composition names. Finally, we prove the insufficiency of open-models, like GPT-4, in achieving the analogous tasks and performances as the proposed domain-specific ones. The work offers a promising avenue for enhancing the effectiveness of descriptor identification in complex physical systems, shedding light over the power of transfer learning to easily adapt and combine available models, with different modalities, to the physics domain, at the same time opening space to a benchmark for LLMs capabilities in such domain.
Current browse context:
cond-mat.mtrl-sci
Change to browse by:
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
IArxiv Recommender
(What is IArxiv?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.