67639 research outputs found
Sort by
Molecular crystal memristors
No full textMemristors have emerged as a promising hardware platform for in-memory computing, but many current devices suffer from channel material degradation during repeated resistive switching. This leads to high energy consumption and limited endurance. Here we introduce a molecular crystal memristor, of which the representative channel material, Sb2O3, possesses a molecular crystal structure where molecular cages are interconnected via van der Waals forces. This unique configuration allows ions to migrate through intermolecular spaces with relatively low energy input, preserving the integrity of the crystal structure even after extensive switching cycles. Our molecular crystal memristor thus exhibits low energy consumption of 26 zJ per operation, with prominent endurance surpassing 109 switching cycles. The device delivers both reconfigurable non-volatile and volatile resistive switching behaviours over a broad range of device scales, from micrometres down to nanometres. Furthermore, we establish the scalability of this technology by fabricating large crossbar arrays on an 8 inch wafer. This enables the successful implementation of reservoir computing on a single CMOS-integrated chip using these memristors, achieving 100% accuracy in dynamic vision recognition.This work was supported by the National Natural Science Foundation of China (22350003, T.Z.; 22535004, T.Z.; U22A20137, Yuan Li; U21A2069, T.Z.) and the National Key R&D Program of China (2021YFA1200501, Yuan Li) and the Interdisciplinary Research Program of HUST (2024JCYJ008, T.Z.) and the Open Research Fund of Suzhou Laboratory (SZLAB-1508-2024-ZD013, T.Z.). We also acknowledge the support of the HPC platform of Huazhong University of Science and Technology
Towards Trustworthy News Recommendation Systems
No full textThe rapid shift to digital news consumption has led to the widespread adoption of personalized news recommendation systems. While these systems enhance content relevance and user engagement, they also present significant challenges related to trustworthiness—particularly in ensuring accuracy, privacy, robustness, and diversity. This thesis aims to strengthen the trustworthiness of news recommendation systems by addressing three critical challenges: the cold start problem (CSP), recommendation unlearning, and filter bubble formation.
To address the cold start problem, we propose GAZRec, a generative adversarial zero-shot learning framework that synthesizes virtual user–news interactions, enabling effective recommendations for new users and articles. Experimental results demonstrate that GAZRec significantly improves accuracy and robustness in cold-start scenarios while maintaining strong performance under standard conditions.
For recommendation unlearning, we introduce FlipRec, a label-flipping-based framework that enables efficient removal of user preferences while preserving model performance. Unlike traditional retraining methods, FlipRec ensures privacy compliance and robustness without incurring high computational costs, making it suitable for large-scale systems.
Additionally, we explore the multi-dimensional nature of filter bubbles by analyzing their formation across topic, sentiment, and political leaning dimensions. Unlike prior studies that focus on single-dimensional biases, our work shows how biases in one dimension can reinforce those in others, progressively narrowing user exposure. By examining the long-term effects of these bubbles, this research offers insights into how personalization can be balanced with diversity—promoting a more inclusive and equitable news environment.
The findings of this thesis contribute to the development of more trustworthy news recommendation systems by improving accuracy in cold-start scenarios, enhancing privacy through efficient unlearning, and fostering diversity by mitigating multi-dimensional filter bubbles. This work lays the foundation for future advancements in ethical and responsible AI-driven news recommendation
Low-call-rate SNPs and presence-absence variation identified in the rice pan-genome can improve genomic prediction of rice gene bank accessions.
No full textKey messageSubstantial improvements in genomic prediction accuracy for rice gene bank accessions were achieved by incorporating SNPs of low call rate identified in a recently published rice pan-genome. Introduction of useful genetic variation to breeding populations is a key factor in achieving genetic gain in crop breeding. However, identifying donors from genetic diversity stored in gene banks requires extensive phenotyping, which is not feasible for many traits of interest. Genomic prediction (GP) of phenotypic values has been proposed to overcome this phenotyping bottleneck. A key challenge for GP is the identification of appropriate markers representative of genetic variation causal for phenotypes. Here we report on utilizing single nucleotide polymorphisms (SNPs) from the core and dispensable genomes of a rice pan-genome resource comprising 16 reference sequences. Using a published pan-genome graph, we identified SNPs within structural variations of the dispensable genome. In this SNP set, SNPs of low call rate (CR) were common. Presence-absence variation (PAV) of these SNPs was associated with subpopulation structure, indicating that SNP absence reflects on underlying sequence PAV rather than being solely due to technical errors in SNP detection. To incorporate these SNPs in GP models, we employed modified encoding, retaining information of PAV and nucleotide variation by one-hot encoding (OHE). Adding these to SNP matrices increased prediction accuracies of GP for some traits and subpopulations. Improvements could largely be attributed to the inclusion of PAV. Our results show that the traditional approach of applying strict CR filters to SNPs located in the dispensable genome disregards potentially valuable genetic information not in linkage with SNPs of high CR. The proposed strategy provides a straightforward way to enhance GP performance in rice gene bank accessions.The authors gratefully acknowledge the access to the Bonna-HPC cluster of the University of Bonn and the support provided by the HPC@HRZ Team of the University of Bonn.
We thank Dr. Yong Zhou from Prof. Wing’s Lab, Center for Desert Agriculture, KAUST, for the initial provision of SNP sets used in this work. We thank Dr. Matt Shenton, Dr. Yoshiaki Ueda, Prof. Dr. Karl Schmid, Prof. Dr. Hans-Peter-Piepho, and Prof. Dr. Annaliese Mason for lively discussions during the drafting of this work.
Open Access funding enabled and organized by Projekt DEAL. This work has been funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy, EXC-2070-390732324-PhenoRob
60 cm2 perovskite-silicon tandem solar cells with an efficiency of 28.9% by homogeneous passivation
No full textInverted perovskite solar cells face performance limitations due to non-radiative recombination at the perovskite surfaces in devices, including functional layers. Advanced characterization and density functional theory reveal that phosphonic acids passivate perovskite surface defects, while piperazinium chloride mitigates interface recombination by improving energy level alignment, introducing a field effect, and homogenizing the surface. Together, the quasi-Fermi level splitting of the perovskite is homogeneously increased by ca. 100 mV. This enables two-terminal perovskite-on-silicon tandems to achieve a certified open-circuit voltage of 2 V for a 1 cm² device and high performance in excess of 31%. The scalability of the passivation is furthermore demonstrated with homogeneously passivated devices reaching certified efficiencies of 28.9% for an active area of 60 cm².The authors thank Dr. Peiliang Chen from Scenergy and Patrick Wyss for wet chemical processing of the Si wafers, Joël Spitznagel, and Sylvain Dunand for the 1 cm2 Si bottom cell fabrications, Julien Gay for the SiOx-np supply, Adrien Theytaz and Jean-David Decoppet for SnOx atomic layer deposition and screen-printing for the large area tandems, Antoine Descoeudres, Vanessa Gainche, and Bertrand Paviet-Salomon for the fabrication of 4 and 60 cm2 Si bottom cells, Gabriel Christmann for PL imaging of 60 cm2 devices, Pascal Alexander Schouwink for GIWAXS measurements and analysis, and Jeong Kwon for supporting the development of large-area tandems. The authors acknowledge funding from the European Union’s Horizon 2020 and innovation program (VIPERLAB, 101006715), the European Commission and the Swiss State Secretariat for Education Research and Innovation (SERI) (TRIUMPH - 101075725 and PEPPERONI - 101084251), the Swiss National Science Foundation (PAPET, 200021_197006; A3P, 40B2-0_1203626, Radicals, CRSII5_216647), the Swiss Federal Office of Energy (PRESTO, PERSISTARS, BESTOBOT), (COMET, 502791-01) and the ETH Domain through an AM grant (AMYS). M.O., D.T., and A.K. acknowledge funding from the European Union’s Horizon 2020 research and innovation program under a Marie Skłodowska-Curie grants (945363 and 101034260). D.T. acknowledges the State Secretariat for Education, Research, and Innovation for an FCS/ESKAS Swiss Government Excellence Scholarship. F.L. and A.F.C.M. thank the BMBF for funding (03EE1183C). F.L. thanks the Volkswagen Foundation for funding via the Freigeist Program. The research reported in this publication was supported by funding from King Abdullah University of Science and Technology (KAUST). For computer time, this research used Shaheen III and Ibex, managed by the Supercomputing Core Laboratory at KAUST
A highly scalable numerical framework for reservoir simulation on UG4 platform
No full textThe modeling and simulation of multiphase fluid flow have received significant attention from the reservoir engineering research community. Many time discretization schemes for multiphase flow equations are either explicit or semi-implicit, relying on the decoupling between the saturation equation and the pressure equation. In this study, we delve into a fully coupled and fully implicit framework for simulating multiphase flow in heterogeneous porous media, considering both the gravity and capillary effects. We utilize the Vertex-Centered Finite Volume Method for spatial discretization and propose an efficient implementation of capillary barrier condition for heterogeneous porous media within the current scheme. Notably, we introduce the Linearly Implicit Extrapolation Method (LIMEX) with an error estimator, adapted for the first time to multiphase flow problems. To solve the resulting linear system, we employ the BiCGSTAB method with the Geometric Multigrid (GMG) preconditioner. The implementations of the models and methods are based on the open-source software: UG4. The results from parallel computations on the supercomputer demonstrate that the scalability of our proposed framework is sufficient, supporting a scale of thousands of processors with Degrees of Freedom (DoF) extending into the billions.This work has been supported by the KAUST Circular Carbon Initiative (CCI), Saudi Arabia project no. REI/1/5217-01-01
Methods for Brain Connectivity Analysis with Applications to Rat Local Field Potential Recordings
No full textModeling the brain dependence network is central to understanding underlying neural mechanisms such as perception, action, and memory. In this study, we present a broad range of statistical methods for analyzing dependence in a brain network. Leveraging a combination of classical and cutting-edge approaches, we analyze multivariate hippocampal local field potential (LFP) time series data concentrating on the encoding of nonspatial olfactory information in rats. We present the strengths and limitations of each method in capturing neural dynamics and connectivity. Our analysis begins with exploratory techniques, including correlation, partial correlation, spectral matrices, and coherence, to establish foundational connectivity insights. We then investigate advanced methods such as Granger causality (GC), robust canonical coherence analysis, spectral transfer entropy (STE), and wavelet coherence to capture dynamic and nonlinear interactions. Additionally, we investigate the utility of topological data analysis (TDA) to extract multi-scale topological features and explore deep learning-based canonical correlation frameworks for connectivity modeling. This comprehensive approach offers an introduction to the state-of-the-art techniques for the analysis of dependence networks, emphasizing the unique strengths of various methodologies, addressing computational challenges, and paving the way for future research.The authors gratefully acknowledge the support and resources provided by the King Abdullah University of Science and Technology (KAUST)
LDM-Based Communication and Computation Co-Design in Integrated Satellite and Aerial Networks
No full textThis paper investigates a highly spectrally efficient transmission scheme in an integrated satellite and aerial network (ISAN). Specifically, we first propose a novel uplink access framework, where the co-design of communication and over-the-air computation (AirComp) is implemented through layer division multiplexing (LDM) in the aerial network, while the cognitive radio-inspired non-orthogonal multiple access (CR-NOMA) technology is employed in the satellite network. Then, according to the proposed framework, we mathematically formulate a joint optimization problem that aims at maximizing the system achievable sum rate, subject to the constraints of minimal accuracy requirement of AirComp and minimal quality-of-service requirements of communication service. Next, by introducing the inter-network interference-related auxiliary variable, we divide the original optimization problem into two subproblems associated with the optimization of the satellite and aerial networks. To tackle the first subproblem, we propose a beamspace-inspired analog beamforming (BF) method, and derive closed-form expressions for BF vectors and transmit powers to implement the CR-NOMA scheme in the satellite network. Meanwhile, to address the second subproblem, we propose a beamspace-inspired digital BF together with successive convex approximation and alternating optimization approaches, to obtain the BF matrices, transmit power coefficients and AirComp scaling factor, so that the LDM-based communication and computation co-design (CCCD) can be realized in the aerial network. Moreover, for complexity reduction, we propose a beamspace-inspired zero-forcing BF method to calculate the communication BF matrices, and then leverage the orthogonal beam superposition approach to obtain the computation BF matrix, thereby presenting another CCCD scheme. Finally, our simulation results confirm that since the proposed schemes can realize spectrum multiplexing for communication and AirComp services, we ach...This work was supported by the National Natural Science Foundation of China under Grant 62471255.(Corresponding author: Min Lin.
Hydrogel Microparticle Encapsulation Enhances In Vivo Peptide Delivery of Recombinant Probiotics
No full textProbiotics, recognized for their cost-effective production, easy storage, and straightforward delivery, are a key focus in developing innovative delivery systems for disease treatment. However, the limited delivery efficacy of probiotics and potential safety issues have hastened their broader application. In this study, a hydrogel encapsulation system is developed to improve the therapeutic efficiency of probiotics through subcutaneous administration. This encapsulation approach is proven to protect the probiotics, thus prolonging their lifetime and enhancing delivery efficiency. Besides, the growth of probiotics is limited within the hydrogel microparticle, which may reduce the adverse effects on the surrounding tissues. This probiotic therapy regimen is demonstrated in Exendin-4 delivery for diabetic treatment, as well as the delivery of thrombopoietin mimetic peptide for thrombocytopenia treatment. In summary, this system has shown the promising potential of probiotic encapsulation in improving the efficacy and safety of probiotic therapy in various diseases.S.G. and K.L. contributed equally to this work. The authors gratefully acknowledge the support from the National Natural Science Foundation of China (82125018, 32430058)
Bayesian Framework for Target Detection in Noise Environments of FDA-MIMO Radar
No full textFor the target detection problem in frequency diverse array multiple input multiple output (FDA-MIMO) radar operating in noise environments (including receiver thermal noise, interference, and clutter after range compensation) with certain prior knowledge available, this paper integrates both training data and prior information of the covariance matrix into the detector design by resorting to Bayesian framework. To model the prior information of the unknown covariance matrix, it is reasonable to assume that it follows a known complex Wishart distribution. Next, this paper designs corresponding detectors within the Bayesian framework based on the one-step and two-step general likelihood ratio test (GLRT) criteria, denoted as OGLRT and TGLRT, as well as the Rao criterion. From the statistical expressions of the detectors, it is evident that the incorporation of prior information about the covariance matrix eliminates the dependency between the number of transmission snapshots and the dimension of the received signals. Specifically, when the number of transmission snapshots is limited, the prior information of the covariance matrix plays a dominant role. Conversely, as the number of transmission snapshots increases, the sample covariance matrix becomes more influential. Simulation results demonstrate that the detectors proposed in this paper exhibit constant false alarm rate (CFAR) characteristics under the null hypothesis. Moreover, even with limited training data, the Bayesian detectors introduced in this paper are still capable of effectively detecting targets and maintain strong robustness against signal mismatches