A5024162917- Advanced Antenna and Metasurface Technologies
- Metamaterials and Metasurfaces Applications
- Antenna Design and Analysis
- Advanced Neural Network Applications
- Acoustic Wave Phenomena Research
- Tensor decomposition and applications
- Parallel Computing and Optimization Techniques
- Probabilistic and Robust Engineering Design
- Indigenous and Place-Based Education
- Coastal wetland ecosystem dynamics
- Software System Performance and Reliability
- Cell Image Analysis Techniques
- Millimeter-Wave Propagation and Modeling
- Image Enhancement Techniques
- Environmental and Agricultural Sciences
- Family Dynamics and Relationships
- Electrochemical Analysis and Applications
- Model Reduction and Neural Networks
- Regional Development and Environment
- Psychosocial Factors Impacting Youth
- Higher Education Research Studies
- China's Ethnic Minorities and Relations
- Education Systems and Policy
- Advanced Computational Techniques and Applications
- Cognitive Abilities and Testing
Institute of Information Engineering
2023-2024
Chinese Academy of Sciences
2023-2024
China Pharmaceutical University
2024
Lehigh University
2024
University of Chinese Academy of Sciences
2023
Nanjing University of Science and Technology
2019-2023
Beijing Language and Culture University
2022
Durham University
2022
Harbin Institute of Technology
2018
China Agricultural University
2015
We introduce GLM-130B, a bilingual (English and Chinese) pre-trained language model with 130 billion parameters. It is an attempt to open-source 100B-scale at least as good GPT-3 (davinci) unveil how models of such scale can be successfully pre-trained. Over the course this effort, we face numerous unexpected technical engineering challenges, particularly on loss spikes divergence. In paper, training process GLM-130B including its design choices, strategies for both efficiency stability,...
Planar metasurfaces have been applied in several fields. Near-field coupling is typically neglected traditional metasurface designs. A numerical modeling method for macrocells that considers near-field couplings between meta-atoms proposed. deep neural network (DNN) constructed to accurately predict the electromagnetic response from different macrocells. Transfer learning employed reduce number of training datasets. The designed embedded optimization algorithm as an effective surrogate...
This study addressed the question, “What factors do experts perceive as impacting STEM achievement of students in rural schools with predominantly Indigenous students?” A thematic analysis interviews 40 educators a depth experience identified six major themes: holistic education, inclusion local culture highly qualified staff, curriculum and instruction, technology, funding. These themes were interrelated. Holistic education demanded more individualized required staff who could adapt...
In this paper, a design of conformal metasurface on to the cylinders is proposed. A 2 bit meta-atom designed cover full <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$2\pi$</tex> phase range, then an extended far field formula from planar employed together with genetic algorithm achieve pre-determined broadband and broad-angle scattering reduction. Numerical full-wave simulations demonstrate our design.
Software-Defined Networking (SDN) decouples the data plane from control plane, enabling centralized and open programmability of network. OpenFlow flow rules are key carrier for SDN application to configure manage through processing efficiency controller in is critical as it will directly impact instantaneity configuring managing plane. Currently, increases by means multi-threaded parallel processing. However, experiments widely used ONOS, we found a new bottleneck that causes performance...
Three different deep learning models were designed in this paper, to predict the electric fields of single nanoparticles, dimers, and nanoparticle arrays. For prediction error was 4.4%. dimers with strong couplings, a sample self-normalization method proposed, reduced by an order magnitude compared traditional methods. arrays, from 28.8% 5.6% previous work. Numerical tests proved validity proposed models, which have potential applications design nanostructures.
Zostera japonica , as one of the major seagrasses in Yellow River Estuary, plays a critical ecological role, particularly providing habitat for marine organisms, stabilizing sediment, and contributing significantly to carbon sequestration. In recent years, seagrass beds have receded extensively due multiple impacts natural factors human activities. This study investigates complex effects extreme climate events activities on growth mechanisms Estuary using combination field sampling,...
The problem of low fertility rate and sharply increased only child family in China is corelated with the implement policy 1978.There are many concerns about changes quality Chinese people.To explore relationship between identities personality, this paper collects 138 results people's TIPI scores compared FFM by group non-only group.The conclusion there no personality differences characters compulsory implemented families that voluntarily choose one-child births.The development overall social...
Surface Enhanced Raman Scattering (SERS) has become a spectroscopic detection technique widely used in many fields. This paper proposes SERS electric field distribution prediction model based on deep learning, the average relative norm error of test set is 3.6%. It takes about 3 minutes to perform an FDTD simulation for Au nanosphere dimer high-performance server, while using learning ordinary PC only less than 0.003 seconds, which can be as kind fast calculation tool optimization algorithm.
Deep neural networks (DNNs) are of critical use in different domains. To accelerate DNN computation, tensor compilers proposed to generate efficient code on domain-specific accelerators. Existing mainly focus optimizing computation efficiency. However, memory access is becoming a key performance bottleneck because the computational accelerators increasing much faster than performance. The lack direct description and data dependence current compilers' intermediate representation (IR) brings...
This paper proposes a physics-driven deep learning model, which defines the custom loss function based on wave equation. The average relative norm error of test set is 3.88%, while it would be 6.36% with standard MSE function. Compared training only using as function, accuracy physically driven model enhanced, and improves convergence rate, especially at early training, well electric field figure quality.
In this paper, we will have a plenty review of our recently proposed nested equivalence source approximation (NESA) for realistic multiscale simulations. the low frequency region, standard NESA is employed directly with linear complexity, high directional fixed number sources are introduced to obtain O(N log N) N unknowns. To future enhance computation performance mixed-frequency problems, mixed-form introduced. The kernel independent, stable, and easy be coded existing method moments (MoM)....
When considering near couplings, the design and optimization of metasurfaces would consume huge time memory costs. Deep neural networks provide unprecedented computational efficiency accuracy for solving complex electromagnetic calculation problems. In this paper, two deep (DNN) are designed to solve fast problems surface-enhanced Raman scattering (SERS) focusing with metasurface respectively. The numerical results prove effectiveness work in paper.
Chinese engineering team participated in the construction of several water conservancy and hydropower projects areas which have differences climate characteristics from China. Due to a lack hydrological climatic data absence systematic design specifications these areas, considering safety, most designs constructions are based on extreme values with codes. The S Project Guinea established rainfall-runoff model through open-source remote sensing satellite precipitation limited station measured...
Boosting the runtime performance of deep neural networks (DNNs) is critical due to their wide adoption in real-world tasks. Existing approaches optimizing tensor algebra expression a DNN only consider expressions representable by fixed set predefined operators, missing possible optimization opportunities between general expressions. We propose OLLIE, first derivation-based program optimizer. OLLIE optimizes programs leveraging transformations expressions, enabling significantly larger search...
In China, for most researchers, there is a discrepancy between promoting Mandarin and conserving heritage language. The study, by using survey scales interview data collected from randomly selected 40 participants, addresses the question what are factors contributing to dialect maintenance of students on individual, familial, school community level. It was concluded that series correlate with students’ use internal external point view, which considers four different levels: Individual,...
Near couplings will lead to low efficiency electromagnetic waves manipulation of the metasurface. In this paper, a space mapping design framework driven by deep learning is proposed for optimization metasurafce with near meta-atoms. The neural network training predict phase atom accurately, in which process designed meta-atoms are considered. and full wave simulation employed as coarse fine models spacing algorithm. Numerical results verify effectiveness framework.