A5103275673- Topic Modeling
- Natural Language Processing Techniques
- Hydrocarbon exploration and reservoir analysis
- Multimodal Machine Learning Applications
- Seismic Imaging and Inversion Techniques
- Geological and Geophysical Studies
- Machine Learning and Data Classification
- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Advanced Graph Neural Networks
- Neural Networks and Applications
- Bone fractures and treatments
- Image Enhancement Techniques
- Traumatic Brain Injury and Neurovascular Disturbances
- Intelligent Tutoring Systems and Adaptive Learning
- Advanced Computational Techniques and Applications
- Reservoir Engineering and Simulation Methods
- earthquake and tectonic studies
- Metallurgical Processes and Thermodynamics
- Geological Studies and Exploration
- Hydraulic Fracturing and Reservoir Analysis
- Machine Learning and Algorithms
- Metallurgy and Material Forming
- Advanced Image Processing Techniques
- Advanced Text Analysis Techniques
Shantou University
2022-2025
Second Affiliated Hospital of Shantou University Medical College
2023-2025
Shanghai Jiao Tong University
2025
China Three Gorges University
2023-2024
Zhejiang University
2023
Yichang Central People's Hospital
2023
Changzhi Medical College
2021-2023
Chinese Academy of Sciences
2010-2023
Central China Normal University
2018-2023
Institute of Geology and Geophysics
2015-2023
Recently deep learning has been successfully adopted in many applications such as speech recognition and image classification. In this work, we explore the possibility of employing graph clustering. We propose a simple method, which first learns nonlinear embedding original by stacked autoencoder, then runs $k$-means algorithm on to obtain clustering result. show that method solid theoretical foundation, due similarity between autoencoder spectral terms what they actually optimize. Then,...
Machine translation has made rapid advances in recent years. Millions of people are using it today online systems and mobile applications order to communicate across language barriers. The question naturally arises whether such can approach or achieve parity with human translations. In this paper, we first address the problem how define accurately measure translation. We then describe Microsoft's machine system quality its translations on widely used WMT 2017 news task from Chinese English....
Automatic neural architecture design has shown its potential in discovering powerful network architectures. Existing methods, no matter based on reinforcement learning or evolutionary algorithms (EA), conduct search a discrete space, which is highly inefficient. In this paper, we propose simple and efficient method to automatic continuous optimization. We call new approach optimization (NAO). There are three key components our proposed approach: (1) An encoder embeds/maps architectures into...
As a new neural machine translation approach, NonAutoregressive Translation (NAT) has attracted attention recently due to its high efficiency in inference. However, the come at cost of not capturing sequential dependency on target side translation, which causes NAT suffer from two kinds errors: 1) repeated translations (due indistinguishable adjacent decoder hidden states), and 2) incomplete transfer source information via states). In this paper, we propose address these problems by...
In underwater scenes, degraded images caused by wavelength-dependent light absorption and scattering present huge challenges to vision tasks. Underwater image enhancement has attracted much attention due the significance of vision-based applications in marine engineering robotics. Numerous algorithms have been proposed last few years. However, almost all existing approaches focus only on independent images. Considering that photographed same scene usually share similar degradation, related...
Recent studies have shown that reinforcement learning (RL) is an effective approach for improving the performance of neural machine translation (NMT) system. However, due to its instability, successfully RL training challenging, especially in real-world systems where deep models and large datasets are leveraged. In this paper, taking several large-scale tasks as testbeds, we conduct a systematic study on how train better NMT using learning. We provide comprehensive comparison important...
Because of the continuous stream touchscreen apps that are claimed to be educational and increasing use devices in early childhood, considerable attention is being paid effect touchscreens on young children's learning. However, existing empirical findings child samples not consistent. In this meta-analysis we tested overall (0- 5-year-olds) learning performance, as well moderators effect, based 36 articles (79 sizes) involving 4,206 participants. The analysis showed a significant (d = 0.46),...
In this paper, we study a new learning paradigm for Neural Machine Translation (NMT). Instead of maximizing the likelihood human translation as in previous works, minimize distinction between and given by an NMT model. To achieve goal, inspired recent success generative adversarial networks (GANs), employ training architecture name it Adversarial-NMT. Adversarial-NMT, model is assisted adversary, which elaborately designed Convolutional Network (CNN). The goal adversary to differentiate...
In this paper, we explore the possibility of leveraging Residual Networks (ResNet), a powerful structure in constructing extremely deep neural network for image understanding, to improve recurrent networks (RNN) modeling sequential data.We show that sequence classification tasks, incorporating residual connections into structures yields similar accuracy Long Short Term Memory (LSTM) RNN with much fewer model parameters.In addition, propose two novel models which combine best both learning...
Zhuohan Li, Zi Lin, Di He, Fei Tian, Tao Qin, Liwei Wang, Tie-Yan Liu. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.
Neural machine translation usually adopts autoregressive models and suffers from exposure bias as well the consequent error propagation problem. Many previous works have discussed relationship between accuracy drop (i.e., left part of translated sentence is often better than its right in left-to-right decoding models) In this paper, we conduct a series analyses to deeply understand problem get several interesting findings. (1) The role on overstated literature, although it indeed contributes...
Sharing source and target side vocabularies word embeddings has been a popular practice in neural machine translation (briefly, NMT) for similar languages (e.g., English to French or German translation). The success of such wordlevel sharing motivates us move one step further: we consider model-level tie the whole parts encoder decoder an NMT model. We share Transformer (Vaswani et al. 2017), state-of-the-art model, obtain compact model named Tied Transformer. Experimental results...
Machine learning is essentially the sciences of playing with data. An adaptive data selection strategy, enabling to dynamically choose different at various training stages, can reach a more effective model in efficient way. In this paper, we propose deep reinforcement framework, which call \emph{\textbf{N}eural \textbf{D}ata \textbf{F}ilter} (\textbf{NDF}), explore automatic and process. particular, NDF takes advantage neural network adaptively select filter important instances from...
Purpose Suicide is a global concern, especially among young people. prediction models have the potential to make it easier identify patients who are at high risk of suicide, but they very little predictive power when there positive value for suicide mortality. Therefore, aim study uncover factors associated with by self-poisoning and further provide trustworthy nomogram predict poisoned patients. Methods This prospectively enrolled 237 were treated poisoning Fifth Medical Center PLA General...
Teaching is critical to human society: it with teaching that prospective students are educated and civilization can be inherited advanced. A good teacher not only provides his/her qualified materials (e.g., textbooks), but also sets up appropriate learning objectives course projects exams) considering different situations of a student. When comes artificial intelligence, treating machine models as students, the loss functions optimized act perfect counterparts objective set by teacher. In...