Article
Biochemical Research Methods
Xinru Ruan, Changzhi Jiang, Peixuan Lin, Yuan Lin, Juan Liu, Shaohui Huang, Xiangrong Liu
Summary: In this study, we propose a novel method using multi-view self-supervised contrastive learning (MSGCL) for miRNA-disease association prediction. By optimizing the graph structure and utilizing the known association network, we enhance the latent representation of association predictions. Experimental results demonstrate that our method outperforms state-of-the-art methods with an improvement of 2.79% in AUC and 3.20% in AUPR.
BRIEFINGS IN BIOINFORMATICS
(2023)
Article
Computer Science, Artificial Intelligence
Xiao Wang, Daisuke Kihara, Jiebo Luo, Guo-Jun Qi
Summary: The study introduces a new EnAET framework to enhance semi-supervised learning methods with self-supervised information. Experimental results demonstrate that the EnAET framework significantly improves the performance of semi-supervised algorithms, even in scenarios with a limited number of images, and can greatly enhance supervised learning as well.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2021)
Article
Engineering, Electrical & Electronic
Xiaofeng Yang, Fengmao Lv, Fayao Liu, Guosheng Lin
Summary: We propose a self-training approach to train vision language BERTs using unlabeled image data, and achieve competitive or better performances with only 300k unlabeled extra data compared to models trained with 3 million extra image data.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Review
Environmental Sciences
Anzhu Yu, Yujun Quan, Ru Yu, Wenyue Guo, Xin Wang, Danyang Hong, Haodi Zhang, Junming Chen, Qingfeng Hu, Peipei He
Summary: This paper summarizes and reviews the methods and challenges of supervised learning with a small number of labeled samples in semantic segmentation tasks of remote sensing images under deep learning framework. It also involves different training methods such as self-supervised learning, semi-supervised learning, weakly supervised learning, and few-shot methods, as well as the solutions to cross-domain challenges.
Article
Computer Science, Artificial Intelligence
Haohang Xu, Hongkai Xiong, Guo-Jun Qi
Summary: This paper proposes a K-Shot Contrastive Learning (KSCL) method for visual features by investigating sample variations within individual instances through multiple augmentations. The experiment results demonstrate that the proposed method achieves superior performances among unsupervised methods.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Tobias Biegel, Patrick Helm, Nicolas Jourdan, Joachim Metternich
Summary: Self-supervised learning has shown impressive performance in anomaly detection tasks. In this paper, we propose SSMSPC, a novel approach for multivariate statistical in-process control (MSPC) based on self-supervised learning. SSMSPC leverages unsupervised representation learning to detect and localize anomalous behavior in discrete manufacturing processes. We introduce a pretext task called Location + Transformation prediction and apply the Hotelling's T-2 statistic for one-class classification in the downstream task. Our evaluation on real-world CNC-milling datasets demonstrates that SSMSPC outperforms state-of-the-art approaches, achieving AUROC scores of 100% and 99.6% respectively. Additionally, we deploy SSMSPC on a CNC-milling machine to showcase its practical applicability as a process monitoring tool.
JOURNAL OF INTELLIGENT MANUFACTURING
(2023)
Article
Computer Science, Artificial Intelligence
Xiao Wang, Guo-Jun Qi
Summary: Representation learning has advanced significantly with contrastive learning methods, which benefit from carefully designed data augmentations maintaining identity. However, these limitations prevent exploration of novel patterns from other transformations. Direct contrastive learning for stronger augmented images proves ineffective. Hence, a general framework called CLSA is proposed, utilizing distribution divergence between weakly and strongly augmented images to supervise retrieval. Experiments on ImageNet and downstream datasets demonstrate significantly boosted performance using information from strongly augmented images, achieving comparable accuracy to supervised results.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Physics, Multidisciplinary
Chi-Ting Ho, Daw-Wei Wang
Summary: The study introduces a self-supervised machine learning method to identify topological phase transitions, demonstrating its robust application in various exactly solvable models. The approach is based on correlating system parameters with non-local observables in ultracold atom experiments.
NEW JOURNAL OF PHYSICS
(2021)
Article
Computer Science, Artificial Intelligence
Zhengyang Yu, Song Wu, Zhihao Dou, Erwin M. Bakker
Summary: Due to its effectiveness and efficiency, deep hashing approaches are widely used for large-scale visual search. However, generating compact and discriminative hash codes for images associated with multiple semantics is still challenging. This paper proposes a novel deep hashing approach, called SADH, which utilizes a self-supervised network and semantic dictionaries to preserve semantic information and employs an asymmetric learning strategy for multi-label semantic preservation. It also introduces a margin-scalable constraint for precise similarity search and robust hash code generation.
Review
Computer Science, Interdisciplinary Applications
Veenu Rani, Syed Tufael Nabi, Munish Kumar, Ajay Mittal, Krishan Kumar
Summary: Machine learning has made significant advances in image processing. Supervised learning relies on labeled data, while unsupervised learning learns from unlabeled data. Self-supervised learning is a type of unsupervised learning that enhances computer vision tasks. This review article provides an in-depth exploration of self-supervised learning and its applications, discussing terms, learning types, and challenges encountered in the process.
ARCHIVES OF COMPUTATIONAL METHODS IN ENGINEERING
(2023)
Article
Computer Science, Information Systems
Meihui Zhong, Mingwei Lin, Zhu He
Summary: In this paper, a novel dynamic multi-scale topological representation (DMTR) method is proposed to improve network intrusion detection performance. The DMTR method achieves the perception of multi-scale topology and exhibits strong robustness. It provides accurate and stable representations even in the presence of data distribution shifts and class imbalance problems. The feasibility and effectiveness of the proposed DMTR method in handling class imbalanced and highly dynamic network traffic are demonstrated through experiments on four publicly available network traffic datasets.
COMPUTERS & SECURITY
(2023)
Article
Computer Science, Artificial Intelligence
Huifang Li, Yidong Li, Yi Jin, Tao Wang
Summary: This study proposes a self-supervised colocalization method that weakens background distraction and mines complementary object regions through object representation enhancement and masked self-supervised learning. The proposed model achieves superior results.
INTERNATIONAL JOURNAL OF INTELLIGENT SYSTEMS
(2022)
Article
Environmental Sciences
Aurelien Colin, Ronan Fablet, Pierre Tandeo, Romain Husson, Charles Peureux, Nicolas Longepe, Alexis Mouche
Summary: This paper studies the semantic segmentation of various features in the oceans using Synthetic Aperture Radar data from satellites. It compares the performance of weakly-supervised and fully-supervised frameworks in identifying different metoceanic processes and finds that the fully-supervised model outperforms the weakly-supervised algorithms.
Article
Computer Science, Artificial Intelligence
Simou Li, Yuxing Mao, Jian Li, Yihang Xu, Jinsen Li, Xueshuo Chen, Siyang Liu, Xianping Zhao
Summary: Self-supervised learning learns valuable representations from unlabeled data, and Federated Self-supervised Learning (FedSSL) combines self-supervised learning with Federated learning to address the privacy concerns of decentralized unlabeled data. The FedUTN framework allows each terminal to train a model that performs well on both IID and non-IID data, and incorporates an aggregation strategy for parameter updates.
APPLIED INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Alexander J. Dyer, Lewis D. Griffin
Summary: Inferring the connectivity of biological neural networks from neural activation data is challenging, but studying the analogous problem in artificial neural networks can provide insights into the biological case. This study focuses on assigning artificial neurons to locations in the LeNet image classifier. A supervised learning approach based on features derived from the activation correlation matrix is evaluated. The experiments suggest that an image dataset needs to fully activate the network and have minimal confounding correlations for accurate localization, and perfect assignment can be achieved by combining features from multiple image datasets.
Review
Materials Science, Multidisciplinary
Chaitanya S. Deo, Elton Y. Chen, Remi Dingeville
Summary: This review explores the application of atomistic modeling techniques in simulating radiation damage in crystalline materials. The formation of defects is a result of radiation damage caused by energetic particles. The subsequent evolution of these defects at various length and time scales requires the use of different simulation techniques to model their diverse behaviors. This work focuses on current and new methodologies at the atomistic scale in investigating the mechanisms of defect formation at the primary damage state.
MODELLING AND SIMULATION IN MATERIALS SCIENCE AND ENGINEERING
(2022)
Article
Materials Science, Multidisciplinary
Alejandro Barrios, James E. Nathaniel, Joseph Monti, Zachary Milne, David P. Adams, Khalid Hattar, Douglas L. Medlin, Remi Dingreville, Brad L. Boyce
Summary: In this study, a new methodology for the fabrication of gradient nanostructured metals via compositional means was developed. By controlling both the alloying elements and the spatially graded grain size distributions, the microstructural stability and ductility of nanocrystalline metals were improved. This fabrication method offers an alternative approach for developing the next generation of microstructurally stable gradient nanostructured films.
Article
Materials Science, Multidisciplinary
Joseph M. Monti, James A. Stewart, Joyce O. Custer, David P. Adams, Diederik Depla, Remi Dingreville
Summary: We propose a generalized multi-phase-field model for predicting the growth of polycrystalline thin films fabricated by physical vapor deposition. This model considers the explicit transport of atomic species to the substrate and the competing diffusion processes on the surface and in the bulk of the film, leading to the formation of films with specific microstructures. Using Monte Carlo simulations with the SiMTRA code, we calculate the energy and direction of arriving atoms at the substrate under magnetron sputtering conditions. Our simulation results agree with the transmission electron microscopy characterization of sputtered films and provide insights into the complex relationships between deposition conditions and bulk and surface morphologies.
Article
Chemistry, Physical
Remi Dingreville, Daniel Vizoso, Ghatu Subhash, Krishna Rajan
Summary: Vibrational spectroscopy is a nondestructive technique used in chemical and physical analyses to determine atomic structures and properties, but evaluating and interpreting spectroscopic profiles based on identifiable peaks can be difficult. To address this challenge, we present a reliable protocol based on supervised manifold learning techniques to connect vibrational spectra to diverse atomic structure configurations.
CHEMISTRY OF MATERIALS
(2023)
Article
Nanoscience & Nanotechnology
Jacob Startt, Mohammed Quazi, Pallavi Sharma, Irma Vazquez, Aseem Poudyal, Nathan Jackson, Remi Dingreville
Summary: The increasing demand for high-performance piezoelectric materials has led to a search for better alternatives to the widely used lead zirconate titanates (PZT) due to their toxicity and thermal stability issues. Doping aluminum nitride (AlN) with scandium (Sc) significantly improves its piezoelectric response. However, the high cost and challenges in fabricating stable films with rare-earth dopants limit their industrial applications. This study combines ab initio calculations with fabrication and experimentation to identify earth-abundant dopants for AlN, and finds that titanium, zirconium, and hafnium induce large piezoelectric enhancements comparable to Sc. This work provides a sustainable and affordable path for the development of next-generation electronics.
ADVANCED ELECTRONIC MATERIALS
(2023)
Article
Materials Science, Multidisciplinary
Yu Yao, Chao Cao, Stephan Haas, Mahak Agarwal, Divyam Khanna, Marcin Abram
Summary: We propose an efficient training framework for machine learning-based emulators and demonstrate its capability to predict the evolution of quantum wave packets. This framework, based on knowledge distillation and curriculum learning, utilizes a set of simple yet physics-rich training examples to teach the emulator the general rules of quantum dynamics. We show that the emulator can learn from these examples and generalize its knowledge to solve more complex cases, while also uncovering new facts and properties of the underlying physical system.
FRONTIERS IN MATERIALS
(2023)
Article
Materials Science, Multidisciplinary
Chongze Hu, Stephane Berbenni, Douglas L. Medlin, Remi Dingreville
Summary: Twinning is a common deformation mechanism in nanocrystalline metals, and solute segregation at twin boundaries plays a vital role in their stability and strengthening. This study reveals a possible discontinuity of solute segregation patterns across a disconnection defect in a wide range of binary alloys. The change in segregation pattern is attributed to the break of local symmetry caused by the disconnection terraces. These findings enhance our understanding of interface segregation phenomena and emphasize the importance of interfacial defects in alloy design.
Article
Chemistry, Physical
Scott Monismith, Jianmin Qu, Remi Dingreville
Summary: In this paper, phase-field simulations are used to investigate fracture and short circuit issues in the Li7La3Zr2O12 (LLZO) solid electrolyte. The study reveals that in the presence of a single crack, the crack propagation threshold exhibits inverse square root scaling with respect to crack length, while the short-circuit potential scales linearly with crack length. For multiple cracks, failure follows the Weibull model, and higher crack density favors failure at lower overpotentials. Additionally, the use of flawless interfacial buffers mitigates failure and allows for larger sustained currents without reaching unstable overpotentials.
JOURNAL OF POWER SOURCES
(2023)
Article
Nanoscience & Nanotechnology
Daniel Vizoso, Chaitanya Deo, Remi Dingreville
Summary: Phase transformations in nanowires induced by external stimuli can be stable under tension and reversible under compressive strain. The recovery of these phase transformations under tensile strain is possible. The stability, reversibility, and recovery of this phase transformation are related to the yielding mechanism and the residual stacking faults and twinning defects.
SCRIPTA MATERIALIA
(2023)
Article
Chemistry, Physical
Dongil Shin, Ryan Alberdi, Ricardo A. Lebensohn, Remi Dingreville
Summary: Recent developments in micromechanics and neural networks have provided promising paths for accurately predicting the response of heterogeneous materials. The deep material network, with its multi-layer design and trained micromechanics building blocks, offers the ability to extrapolate material behavior to different constitutive laws without retraining. However, the random initialization of network parameters in current training methods leads to unavoidable errors. In this study, we propose a visualization technique using an analogous unit cell to initialize deeper networks and improve the accuracy and calibration performance, while also providing a more intuitive representation of the network for explainability.
NPJ COMPUTATIONAL MATERIALS
(2023)
Article
Materials Science, Multidisciplinary
Saaketh Desai, Ankit Shrivastava, Marta D'Elia, Habib N. Najm, Remi Dingreville
Summary: This study investigates and discusses the ability of various linear and nonlinear dimensionality reduction methods to quantify and characterize microstructure evolution, providing considerations and guidelines for choosing dimensionality reduction methods in materials problems involving high dimensional data.
Review
Materials Science, Multidisciplinary
Chongze Hu, Remi Dingreville, Brad L. Boyce
Summary: Most materials, including metals, alloys, ceramics, and composites, are polycrystalline and have grain boundaries that affect their properties. The presence of intentional solutes or impurities near the grain boundaries can influence their behavior and stability. Advanced electron microscopy techniques allow researchers to directly observe grain boundary structures and segregation. However, computational modeling techniques are indispensable for understanding the underlying mechanisms of grain boundary segregation.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Chemistry, Physical
Ryan M. Khan, Martin Rejhon, Yanxiao Li, Nitika Parashar, Elisa Riedo, Ryan R. Wixom, Frank W. DelRio, Remi Dingreville
Summary: As the field of low-dimensional materials continues to grow, there is a need for techniques to characterize the mechanical properties of these materials at the nanoscale. This paper presents a modulated nanomechanical measurement technique based on atomic-force microscopy, which enables non-destructive measurements of the elasticity of ultra-thin materials with high resolution. The technique is used to study the stiffness dependence of graphene thin films and discover a peak transverse modulus in two-layer graphene.
Article
Materials Science, Multidisciplinary
James A. Stewart, Jacob K. Startt, Remi Dingreville
Summary: Through atomistic simulations, this study investigates the dynamic properties and equation-of-state of the Cantor alloy under shock-loading conditions. The role of local phase transformations and the alloy's high spall strength are revealed. The results validate the predictability of the model and provide insights for further advancements in applications of this alloy under extreme environments.
MATERIALS RESEARCH LETTERS
(2023)
Article
Materials Science, Multidisciplinary
Timothy A. Elmslie, Jacob Startt, Yang Yang, Sujeily Soto-Medina, Emma Zappala, Mark W. Meisel, Michele Manuel, Benjamin A. Frandsen, Remi Dingreville, James J. Hamlin
Summary: Magnetic properties of Cantor alloy samples with varying composition were investigated using magnetometry and muon spin relaxation. Two transitions were observed: a spin-glass-like transition between 55 K and 190 K depending on composition, and a ferrimagnetic transition at approximately 43 K in multiple samples. The magnetic signatures at 43 K were not affected by chemical composition. The effective magnetic moment decreased with increasing Cr or Mn concentrations and increased with decreasing Fe, Co, or Ni concentrations. The results provide insights into controllable tuning of the magnetic properties of Cantor alloy variants.