Article
Chemistry, Multidisciplinary
Daiguo Deng, Zengrong Lei, Xiaobin Hong, Ruochi Zhang, Fengfeng Zhou
Summary: This study used a heterogeneous graph neural network (MolHGT) to represent molecular structures, which showed improved performance in molecular property predictions compared to existing studies.
Article
Chemistry, Analytical
Haolin Zhan, Xin Zhu, Zhiwei Qiao, Jianming Hu
Summary: Determining various properties of molecules is critical in drug discovery. However, the poor interpretability of deep learning models reduces their credibility. Therefore, this study develops a novel framework that utilizes graph neural networks for feature extraction, curriculum-based learning strategies for optimization, and a Learning Binary Neural Tree (LBNT) for prediction, to improve the performance of neural networks and reveal their decision-making process to chemists.
ANALYTICA CHIMICA ACTA
(2023)
Article
Crystallography
Chao Shu, Junjie He, Guangjie Xue, Cheng Xie
Summary: This paper presents a method based on grain knowledge graph representation learning to predict the properties of polycrystalline materials. By constructing an advanced digital representation of the knowledge graph and proposing a heterogeneous grain graph attention model, it achieves feature embedding of the microstructure and mining of the relationship between the structure and the material properties.
Article
Computer Science, Theory & Methods
Guanghan Duan, Hongwu Lv, Huiqiang Wang, Guangsheng Feng
Summary: Deep learning greatly enhances binary anomaly detection capabilities, but the performance in intrusion class differentiation is still insufficient. Two challenges, emphasizing statistical attack characteristics and the need for high-quality labeled data samples, have not been fully explored. To address these issues, a dynamic line graph neural network (DLGNN)-based intrusion detection method with semisupervised learning is proposed. Experimental results on 6 novel datasets demonstrate high accuracy in abnormality detection and state-of-the-art multiclass performance.
IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY
(2023)
Article
Electrochemistry
Danpeng Cheng, Wuxin Sha, Qigao Han, Shun Tang, Jun Zhong, Jinqiao Du, Jie Tian, Yuan-Cheng Cao
Summary: LiNixCoyMn1-x-yO2 (NCM) is a critical cathode material for lithium-ion batteries in electric vehicles. The aging of cathode/electrolyte interfaces leads to capacity degradation and long-term cycle instability. A novel neural network model called ACGNet is developed to predict electrochemical stability windows of crystals, allowing for high-throughput screening of coating materials. LiPO3 is identified as a promising coating material with high oxidation voltage and low cost, which significantly improves the cycle stability of NCM batteries. This study demonstrates the accuracy and potential of machine learning in battery materials.
ELECTROCHIMICA ACTA
(2024)
Article
Engineering, Multidisciplinary
Asad Khan, Sakander Hayat, Yubin Zhong, Amina Arif, Laiq Zada, Meie Fang
Summary: A neural network is a computer system that imitates nerve tissue and the nervous system. It has various applications in fields such as computer graphics, AI, machine learning, chemistry, and material science. Researchers have recently started studying the structural properties of neural networks using mathematical tools and graph theory. They have analyzed different types of neural networks and their graph structures, reporting results on clique number, chromatic number, independence number, matching ratio, and domination number.
ALEXANDRIA ENGINEERING JOURNAL
(2023)
Article
Computer Science, Artificial Intelligence
Eric Paquet, Farzan Soleymani
Summary: This paper introduces a new hybrid deep quantum neural network, named QuantumLeap system, for financial predictions. By transforming financial time series into a sequence of density matrices and combining with the measurement from classical network, this system is able to predict the maximum price of securities accurately and efficiently.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Computer Science, Interdisciplinary Applications
Xuefen Lin, Jielin Chen, Weifeng Ma, Wei Tang, Yuchen Wang
Summary: This study proposes an improved graph convolution model that achieves effective emotion classification in complex dataset environments and reduces the cost of affective computing.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2023)
Review
Electrochemistry
Yunqi Shao, Lisanne Knijff, Florian M. Dietrich, Kersti Hermansson, Chao Zhang
Summary: Batteries and supercapacitors are electrochemical energy storage systems that require a molecular-level understanding of electrolytes for optimizing performance and ensuring safety. Atomistic machine learning is a promising technology for bridging microscopic models and macroscopic phenomena.
BATTERIES & SUPERCAPS
(2021)
Article
Automation & Control Systems
Enjamamul Hoq, Osama Aljarrah, Jun Li, Jing Bi, Alfa Heryudono, Wenzhen Huang
Summary: This article explores different methods for predicting full stress fields in random heterogeneous materials, including model order reduction with classical machine learning and computer vision-based deep learning. The study finds that deep learning methods provide more accurate predictions with reduced errors compared to classical machine learning techniques.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2023)
Article
Computer Science, Information Systems
Gabriel St-Pierre Lemieux, Eric Paquet, Herna L. Viktor, Wojtek Michalowski
Summary: This work introduces novel approaches based on geometrical deep learning to predict protein-protein interactions. A dataset from the Negatome Database, containing both interacting and non-interacting proteins, is used. Interactions are predicted from a graph representation of the proteins' 3D macromolecular surfaces, with nodes described using heat and wave kernel signatures. Twenty-one neural network architectures, including graph convolutional neural networks, spectral convolutional neural networks, and a novel spatio-spectral spatialized-gated convolutional neural network, are proposed and compared. Experimental results demonstrate the accuracy and efficiency of the proposed architectures.
Article
Computer Science, Information Systems
Liwei Jiang, Guanghui Yan, Hao Luo, Wenwen Chang
Summary: This paper proposes an improved collaborative filtering recommendation model that addresses the noise and sparsity issues in knowledge graphs, resulting in improved recommendation accuracy and effectiveness through enhanced embedding quality and personalization capability.
Article
Computer Science, Artificial Intelligence
Ram Krishn Mishra, Siddhaling Urolagin, J. Angel Arul Jothi, Pramod Gaur
Summary: Image processing is a technique used to apply various operations to images to improve them or extract information, with facial recognition being a prominent application. This study examines the accuracy of categorizing human facial expressions using deep learning and transfer learning methods, proposing a deep hybrid learning approach that combines multiple deep learning models.
IMAGE AND VISION COMPUTING
(2022)
Article
Materials Science, Multidisciplinary
Prathik R. Kaundinya, Kamal Choudhary, Surya R. Kalidindi
Summary: Machine learning has greatly enhanced traditional materials discovery and design pipeline, particularly in predicting material properties. However, predicting complex spectral targets such as electron density of states (DOS) remains challenging. This study presents an extension of the atomistic line graph neural network to accurately predict DOS and evaluates two methods of target representation.
Article
Computer Science, Artificial Intelligence
Yunan Luo, Yang Liu, Jian Peng
Summary: The paper introduces KDBNet, a deep learning algorithm that incorporates 3D structure data of proteins and molecules to predict binding affinities in protein kinases. The algorithm utilizes graph neural networks to capture the geometric and spatial characteristics of binding activity. The study demonstrates that KDBNet outperforms existing deep learning models and its predictions are calibrated with respect to uncertainty. When integrated with Bayesian optimization, KDBNet enables efficient active learning and accelerates the exploration of kinase-drug pairs.
NATURE MACHINE INTELLIGENCE
(2023)
Article
Chemistry, Physical
Alejandro Rodriguez, Changpeng Lin, Hongao Yang, Mohammed Al-Fahdi, Chen Shen, Kamal Choudhary, Yong Zhao, Jianjun Hu, Bingyang Cao, Hongbin Zhang, Ming Hu
Summary: Existing machine learning methods for predicting phonon properties of crystals are limited due to the complexity scaling with the number of atomic species. In this study, we address this issue by developing the Elemental Spatial Density Neural Network Force Field (Elemental-SDNNFF) and demonstrate its effectiveness on predicting phonon properties of various Heusler structures. We also gain insights into the ultralow lattice thermal conductivity and discover double Weyl points in different Heusler structures.
NPJ COMPUTATIONAL MATERIALS
(2023)
Article
Chemistry, Medicinal
Vishu Gupta, Kamal Choudhary, Yuwei Mao, Kewei Wang, Francesca Tavazza, Carelyn Campbell, Wei-keng Liao, Alok Choudhary, Ankit Agrawal
Summary: The applications of artificial intelligence, machine learning, and deep learning techniques in materials science have become increasingly common. Predictive models based on deep transfer learning are deployed in an online software tool to accelerate materials discovery and design. The tool can generate up to 41 different material property values based on the input material compositions.
JOURNAL OF CHEMICAL INFORMATION AND MODELING
(2023)
Article
Chemistry, Medicinal
Kamal Choudhary, Ramya Gurunathan, Brian DeCost, Adam Biacchi
Summary: This study presents an integrated and general-purpose AtomVision library for generating and curating microscopy image datasets and applying various machine learning techniques. The library is demonstrated through several applications, such as materials classification, pixel-wise classification, and super-resolution.
JOURNAL OF CHEMICAL INFORMATION AND MODELING
(2023)
Article
Chemistry, Physical
Kangming Li, Brian DeCost, Kamal Choudhary, Michael Greenwood, Jason Hattrick-Simpers
Summary: Recent advances in machine learning have improved performance in material database benchmarks, but good benchmark scores may not guarantee good generalization performance. ML models trained on earlier data can have degraded performance on new compounds due to distribution shift. Simple tools like UMAP and disagreement analysis between ML models can help foresee this issue. Strategies guided by UMAP and query by committee can significantly improve prediction accuracy with minimal additional test data. This work provides valuable insights for building robust and generalizable databases and models.
NPJ COMPUTATIONAL MATERIALS
(2023)
Article
Chemistry, Physical
Yuwei Mao, Mahmudul Hasan, Arindam Paul, Vishu Gupta, Kamal Choudhary, Francesca Tavazza, Wei-keng Liao, Alok Choudhary, Pinar Acar, Ankit Agrawal
Summary: Materials design aims to identify material features that provide optimal properties for engineering applications. This paper proposes an AI-driven microstructure optimization framework for elastic properties of materials, which can discover multiple polycrystalline microstructures without compromising optimal property values. The framework was evaluated for Titanium in JARVIS database and is expected to be widely applicable for materials with multiple crystal systems.
NPJ COMPUTATIONAL MATERIALS
(2023)
Article
Materials Science, Multidisciplinary
Kevin F. Garrity, Kamal Choudhary
Summary: Parametrized tight-binding models can efficiently and accurately predict properties of molecules and solids, but limited availability of well-tested parameter sets hinders routine use. To overcome this, a density functional theory database of nearly 1,000,000 materials is developed to fit a universal set of tight-binding parameters for 65 elements and their binary combinations, including two-body and three-body effective interaction terms. The model allows for metallic, covalent, and ionic bonds with the same parameter set and is continuously improved using a learning framework.
PHYSICAL REVIEW MATERIALS
(2023)
Article
Materials Science, Multidisciplinary
Ramya Gurunathan, Kamal Choudhary, Francesca Tavazza
Summary: This study presents an atomistic line graph neural network (ALIGNN) model for predicting the phonon density of states and related thermal and thermodynamic properties. The model is trained on a database of over 14000 phonon spectra from the JARVIS-DFT database and shows accurate predictions for properties such as heat capacity, vibrational entropy, and isotopic phonon-scattering rate.
PHYSICAL REVIEW MATERIALS
(2023)
Article
Electrochemistry
Runze Zhang, Robert Black, Debashish Sur, Parisa Karimi, Kangming Li, Brian DeCost, John R. Scully, Jason Hattrick-Simpers
Summary: AutoEIS is an open-source tool that automates electrochemical impedance spectroscopy (EIS) analysis by proposing statistically plausible equivalent circuit models (ECMs). It demonstrates generalizability by successfully analyzing EIS datasets from three distinct electrochemical systems. AutoEIS identifies competitive or superior ECMs to those recommended by experts, providing statistical indicators of the preferred solution.
JOURNAL OF THE ELECTROCHEMICAL SOCIETY
(2023)
Article
Chemistry, Medicinal
Kamal Choudhary, Ramya Gurunathan, Brian DeCost, Adam Biacchi
Summary: This article introduces an integrated and general-purpose AtomVision library for generating and curating microscopy image datasets and applying machine learning techniques. The authors demonstrate the library's applicability through various experiments and models on material design applications.
JOURNAL OF CHEMICAL INFORMATION AND MODELING
(2023)
Article
Chemistry, Multidisciplinary
Kamal Choudhary, Brian Decost, Lily Major, Keith Butler, Jeyan Thiyagalingam, Francesca Tavazza
Summary: Classical force fields based on machine learning methods have shown great potential for large scale simulations of solids. However, existing models are usually limited to specific systems. In this study, a unified atomistic line graph neural network-based force field model is developed, which can model structurally and chemically diverse solids with any combination of 89 elements from the periodic table. The model is trained and validated using a large dataset, demonstrating its accuracy and applicability.
Article
Materials Science, Multidisciplinary
Howie Joress, Bruce Ravel, Elaf Anber, Jonathan Hollenbach, Debashish Sur, Jason Hattrick-Simpers, Mitra L. Taheri, Brian DeCost
Summary: Short-range order (SRO) plays a critical role in the properties of multicomponent alloys, and the use of extended X-ray absorption fine structure (EXAFS) technique for quantitative analysis of SRO is discussed in this article.
Article
Chemistry, Physical
Joshua Ojih, Chen Shen, Alejandro Rodriguez, Hongbin Zhang, Kamal Choudhary, Ming Hu
Summary: This study presents an efficient workflow combining high-throughput density functional theory computing and machine learning models for screening ultralow lattice thermal conductivity from large-scale inorganic crystals. By training multiple machine learning models on large datasets and analyzing the correlation between thermal conductivity and material features, the researchers identified two excellent material descriptors. The workflow offers a new route to accelerate the discovery of materials with ultralow thermal conductivity.
JOURNAL OF MATERIALS CHEMISTRY A
(2023)