A5032074795- Parallel Computing and Optimization Techniques
- Cloud Computing and Resource Management
- Advanced Vision and Imaging
- Distributed and Parallel Computing Systems
- Advanced Graph Neural Networks
- Real-Time Systems Scheduling
- VLSI and Analog Circuit Testing
- Generative Adversarial Networks and Image Synthesis
- Network Security and Intrusion Detection
- Graph Theory and Algorithms
- Speech and Audio Processing
- Integrated Circuits and Semiconductor Failure Analysis
- Sleep and Work-Related Fatigue
- Speech Recognition and Synthesis
- Traffic and Road Safety
- IoT and Edge/Fog Computing
- Smart Grid Security and Resilience
- Power Systems and Technologies
- Ecology and Vegetation Dynamics Studies
- Image Enhancement Techniques
- Vehicular Ad Hoc Networks (VANETs)
- Human-Automation Interaction and Safety
- Network Time Synchronization Technologies
- Scheduling and Optimization Algorithms
- 3D Shape Modeling and Analysis
Hunan University of Science and Technology
2022-2024
Southwest Petroleum University
2024
Hunan University
2011-2023
Sichuan University
2023
West China Hospital of Sichuan University
2023
State Key Laboratory of Chemobiosensing and Chemometrics
2023
Yunnan University
2023
Air Force General Hospital PLA
2019-2023
Shanghai Jiao Tong University
2021
State Grid Hunan Electric Power Company Limited
2021
AI is undergoing a paradigm shift with the rise of models (e.g., BERT, DALL-E, GPT-3) that are trained on broad data at scale and adaptable to wide range downstream tasks. We call these foundation underscore their critically central yet incomplete character. This report provides thorough account opportunities risks models, ranging from capabilities language, vision, robotics, reasoning, human interaction) technical principles(e.g., model architectures, training procedures, data, systems,...
Hypergraph, an expressive structure with flexibility to model the higher-order correlations among entities, has recently attracted increasing attention from various research domains. Despite success of Graph Neural Networks (GNNs) for graph representation learning, how adapt powerful GNN-variants directly into hypergraphs remains a challenging problem. In this paper, we propose UniGNN, unified framework interpreting message passing process in and hypergraph neural networks, which can...
This paper aims to improve the widely used deep speaker embedding x-vector model. We propose following improvements: (1) a hybrid neural network structure using both time delay (TDNN) and long short-term memory networks (LSTM) generate complementary information at different levels; (2) multi-level pooling strategy collect from TDNN LSTM layers; (3) regularization scheme on extraction layer make extracted embeddings suitable for fusion step. The synergy of these improvements are shown NIST...
Speaker verification systems often degrade significantly when there is a language mismatch between training and testing data. Being able to improve cross-lingual speaker system using unlabeled data can greatly increase the robustness of reduce human labeling costs. In this study, we introduce an unsupervised Adversarial Discriminative Domain Adaptation (ADDA) method effectively learn asymmetric mapping that adapts target domain encoder source domain, where are speech from different...
In this paper, the joint optimization problem with energy efficiency and effective resource utilization is investigated for heterogeneous distributed multi-core embedded systems. The system model considered to be fully a model, that is, all nodes have different maximum speeds power consumption levels from perspective of hardware while they can employ scheduling strategies applications. Since concerned by nature multi-constrained multi-variable in which closed-form solution cannot obtained,...
As an essential part of Internet Things, monocular depth estimation (MDE) predicts dense maps from a single red-green-blue (RGB) image captured by cameras. Past MDE methods almost focus on improving accuracy at the cost increased latency, power consumption, and computational complexity, failing to balance efficiency. Additionally, when speeding up algorithms, researchers commonly ignore their adaptation different hardware architectures edge devices. This article aims solve these challenges....
Multiprocessor systems are increasingly deployed in real-time applications, where reliability, energy consumption, and makespan often the main scheduling objectives. In this work, we investigate dynamic of tasks modeled by directed acyclic graphs (DAGs), which is an NP-hard problem with all existing methods being heuristics. Our contributions have two steps: 1) assuming that allocation DAG nodes to processors given, propose optimal (OEA) search-based OEA (SOEA)-the first minimize consumption...
Heterogeneous computing systems are being increasingly deployed on time-critical applications, where tasks need to meet execution deadlines and the energy consumption is be minimized. Dynamic voltage frequency scaling (DVFS) has been widely applied for saving devices. Unfortunately, DVFS may introduce transient errors shorten processor lifetime. There also time overhead when making switching. In this article, we investigate scheduling approaches—that independent of, or weakly dependent...
Interspecific variation in plant functional traits is the basis of species coexistence natural ecosystems. However, intraspecific extremely important to community assemblage as well. Here, we quantify leaf trait within and across 32 dominant shrub two different forms (16 evergreen 16 deciduous species) subtropical broadleaf mixed forest Karst topography Guilin, southwest China. Results showed that area (LA) thickness (LT) were significantly lower than those species, whereas specific areas...
Abstract Hemicyanine dyes, with a tunable optical site and high wavelength tailorability, are of significant importance in the fields sensing diagnosis. Following discovery near-infrared (NIR) (650–900 nm) fluorescent dyes Changsha (CS) Huda (HD) by our group, remarkable progress has been made development hemicyanine-based probes for vivo imaging detecting. In this review, we summarize key contributions group developing long-wavelength (650–1700 hemicyanines utilizing them to construct...
Due to the excellent stability of titanium dioxide (TiO2), there is still value in improving its solar-to-hydrogen conversion efficiency through tremendous attempts. Metal sulfides with a narrow bandgap are good candidates broaden ultraviolet light absorption TiO2 into visible region. However, suffer from photocorrosion issue, leading poor stability. Herein, type-II heterojunction TiO2/In2S3 fabricated by hydrothermal method, and NiFe Prussian blue analog (NFP) overlayer deposited on surface...
<abstract> <p>Estimating the binding affinity between proteins and drugs is very important in application of structure-based drug design. Currently, applying machine learning to build protein-ligand prediction model, which helpful improve performance classical scoring functions, has attracted many scientists' attention. In this paper, we have developed an model called GAT-Score based on graph attention network (GAT). The complex represented by a structure, atoms protein ligand...
Energy provision is known and limited for battery-constrained/powered systems (e.g., mobile phones, electric cars, robots, wireless sensor networks). System-level energy-aware design methodologies have been proposed to control energy usage guarantee, optimize the end-to-end response time high-performance in heterogeneous systems, where pre-assignment strategies are considered as their main contributions. However, when pre-assigning tasks, these either overly pessimistic or adopt an average...
Precise driving status recognition is a prerequisite for human–vehicle collaborative systems towards sustainable road safety. In this study, simulated platform was built to capture multimodal information simultaneously, including vision-modal data representing driver behaviour and sensor-modal vehicle motion. Multisource are used quantify the risk of distracted from four levels, safe driving, slight risk, moderate severe rather than detecting action categories. A fusion method called...
Automotive cyber-physical systems (ACPS) are typical because of the joint interaction between cyber part and physical part. Functional safety requirement (including response time reliability requirements) for an ACPS function must be assured safe driving. Auto industry is cost-sensitive, power-sensitive, environment-friendly. Energy consumption affects development efficiency living environment people. This paper solves problem optimizing energy while assuring its functional (i.e.,...
Depth estimation has received considerable attention and is often applied to visual simultaneous localization mapping (SLAM) for scene reconstruction. At least our knowledge, sufficiently reliable depth always fails be provided monocular estimation-based SLAM because new image features are rarely re-exploited effectively, local easily lost, relative relationships among pixels readily ignored in previous methods. Based on inaccurate estimation, still faces scale ambiguity problems. To...
Optimisation of timing performance and power consumption on heterogeneous multi-core architectures is gaining increasing attention. Systems devices may have varying demands power, which motivates more flexible optimisation. Along this line, we consider a computing architecture with multiple cores, where each core runs mixed stream general dedicated tasks certain scheduling strategy. Employing the queuing model, first propose load balancing algorithm, minimises average response time whilst...
A double Hamming distance‐based 2D reordering method is proposed for a test set with don't care bits ( X s) to reduce scan‐in power and compress patterns. In this method, the rows columns in are reordered sequentially so that similar or aggregated together. Being different from other reordering‐based methods, authors' reorders every two more identical closer whole process, where s clustered. Each then replaced 0 1 using minimum transition filling scheme minimise '1‐0' '0‐1' logical state...
User's power smart meter value is frequently uploaded to the company by meters, which makes users face privacy leakage issues. The sends data through concentrator. So concentrator has a large amount of be transmitted. Therefore, we design secure certificateless aggregation signcryption scheme for this problem. Using not only protects user's usage information from being leaked, but also reduces transmitted aggregation. Through calculation and comparison, it concluded that implementing shorter...