Please help repost and spread the word. 🙏 My research group at @UBuffalo's Materials Design & Innovation (@UBengineering & @UBCAS) is recruiting multiple graduate students! ubwp.buffalo.edu/jiayu-peng-… If you are interested in combining 🤖 data science and machine learning with ⚛️ materials physics and surface chemistry to design materials and interfaces for⚡ decarbonization and 🧪 sustainability, please visit our website and reach out to me by email (see the link above for more details). #sciencetwitter #research #chemistry #materials #machinelearning #artificalintelligence

📢 New preprint alert! Our latest work @RGBLabMIT, “Learning Ordering in Crystalline Materials with Symmetry-Aware Graph Neural Networks,” is out on arXiv: arxiv.org/abs/2409.13851 ⚛️ Chemical ordering in crystalline materials, ranging from well-ordered elemental arrangements to fully disordered solid solutions, can strongly influence their key properties, such as stability, strength, and catalytic performance. The neglect of chemical ordering in AI-driven materials discovery has drawn concern from the experimental community, as the AI-predicted ordered structures might turn out to be disordered in reality. 🤖 Following our previous work (doi.org/10.1016/j.xcrp.2024.…), we systematically benchmarked and contrasted a family of powerful representation learning tools for crystalline materials—graph convolutional neural networks (GCNNs)—regarding their capability for capturing and differentiating various symmetrically inequivalent atomic orderings, and more importantly, learning how materials properties depend on diverse orderings. ✅ We found that symmetry-invariant GCNNs fail catastrophically when learning the ordering dependence of materials properties. In contrast, equivariant GCNNs perform much better due to their inherent ability to preserve and distinguish between the distinct crystallographic symmetries of inequivalent orderings, as shown by their well-structured latent space with well-separated representations learned from different orderings. 🧑‍💻 For method-makers, we clarified and formalized an overlooked standing challenge in the representation learning of crystalline materials, provided a large custom-made dataset with task-specific labels and data splits, and further benchmarked a wide variety of models to lay out what SOTA performance looks like. 👨‍🔬 For method-users, we clearly demonstrated why model architecture matters and offered new guidance on model selection to facilitate more rigorous, ordering-aware AI-powered materials discovery, where the dependence of materials properties on chemical ordering can be well captured and accurately considered. #chemtwitter #sciencetwitter #research #chemistry #materials #machinelearning #artificalintelligence

I am thrilled to share that I will join the Department of Materials Design and Innovation at @UBuffalo as an Assistant Professor in January 2025! I feel very excited to be a part of a forward-leaning #MatSci department uniquely centered on the nexus of materials science and information/data science across @UBengineering and @UBCAS. ⚛️🤖 My research group will focus on understanding and resolving the formidable complexities of materials and interfaces under harsh electrochemical conditions for #Sustainability and #Decarbonization through building #PhysicsBased and #DataDriven #MachineLearning methods that capture/optimize materials thermodynamics and surface kinetics from #AtomisticSimulations and #Electrochemistry. ⚡🧪 #NewPI #UBTrueBlue