A5042217094- Stochastic Gradient Optimization Techniques
- Advanced Graph Neural Networks
- Graph Theory and Algorithms
- Topic Modeling
- Deception detection and forensic psychology
- Occupational Health and Safety Research
- Natural Language Processing Techniques
- Neural Networks and Applications
- Advanced Neural Network Applications
- Semantic Web and Ontologies
- Human Pose and Action Recognition
- Smart Grid and Power Systems
- Educational Innovations and Challenges
- Ferroptosis and cancer prognosis
- Insect symbiosis and bacterial influences
- Advanced Multi-Objective Optimization Algorithms
- Sulfur Compounds in Biology
- Gaussian Processes and Bayesian Inference
- Traffic Prediction and Management Techniques
- Renal cell carcinoma treatment
- Heme Oxygenase-1 and Carbon Monoxide
- Traffic control and management
- Advanced Vision and Imaging
- Translation Studies and Practices
- Autonomous Vehicle Technology and Safety
University of Notre Dame
2020-2024
South China University of Technology
2024
McMaster University
2018-2019
Endometrial cancer (EC) stands as one of the most prevalent gynecological malignancies affecting women, with its incidence and disease-related mortality steadily on rise. Disulfiram (DSF), an FDA-approved medication primarily used for treating alcohol addiction, has exhibited promising anti-tumor properties. Studies have revealed DSF's capacity enhanced activity, particularly when combined copper. The novel Copper-Cysteamine (CuCy) compound, Cu
In this work, we propose to employ information-geometric tools optimize a graph neural network architecture such as the convolutional networks. More specifically, develop optimization algorithms for graph-based semi-supervised learning by employing natural gradient information in process. This allows us efficiently exploit geometry of underlying statistical model or parameter space and inference. To best our knowledge, is first work that has utilized networks can be extended other problems....
A gated branch neural network (GBNN) is proposed for modelling mandatory lane changing (MLC) behaviour at the on‐ramps of highways. It provides a core algorithm an MLC suggestion system advanced driver assistance systems (ADAS), where main challenge trade‐off between computational speed and prediction accuracy both non‐merge merge events. The GBNN employs based on correlation analysis, scaled exponential linear units activation function, adaptive moment estimation optimiser. has been...
We propose extrinsic and intrinsic deep neural network architectures as general frameworks for learning on manifolds. Specifically, networks (eDNNs) preserve geometric features manifolds by utilizing an equivariant embedding from the manifold to its image in Euclidean space. Moreover, (iDNNs) incorporate underlying geometry of via exponential log maps with respect a Riemannian structure. Consequently, we prove that empirical risk minimizers (ERM) eDNNs iDNNs converge optimal rates. Overall,...
The inactivation of von Hippel–Lindau (VHL) is critical for clear cell renal carcinoma (ccRCC) and VHL syndrome. loss leads to the stabilization hypoxia-inducible factor α (HIFα) other substrate proteins, which, together, drive various tumor-promoting pathways. There inadequate molecular characterization restoration in VHL-defective ccRCC cells. identities HIF-independent substrates remain elusive. We reinstalled expression 786-O performed transcriptome, proteome ubiquitome profiling assess...
Open intent detection, a crucial aspect of natural language understanding, involves the identification previously unseen intents in user-generated text. Despite progress made this field, challenges persist handling new combinations components, which is essential for compositional generalization. In paper, we present case study exploring use ChatGPT as data augmentation technique to enhance generalization open detection tasks. We begin by discussing limitations existing benchmarks evaluating...
With the widespread adoption of large language models (LLMs) in numerous applications, challenge factuality and propensity for hallucinations raises significant concerns. To address this issue, particularly retrieval-augmented in-context learning, we introduce hierarchical graph thoughts (HGOT), a structured, multi-layered approach designed to enhance retrieval pertinent passages during learning. The framework utilizes emergent planning capabilities LLMs, employing divide-and-conquer...
With significant increases in wireless link capacity, edge devices are more connected than ever, which makes possible forming artificial neural network (ANN) federations on the devices. Partition is key to success of distributed ANN inference while unsolved because unclear knowledge representation most models. We propose a novel partition approach (TeamNet) based psychologically-plausible competitive and selective learning schemes evaluating its performance carefully with thorough...
The success of deep neural networks (DNN) in machine perception applications such as image classification and speech recognition comes at the cost high computation storage complexity. Inference uncompressed large scale DNN models can only run cloud with extra communication latency back forth between end devices, while compressed achieve real-time inference on devices price lower predictive accuracy. In order to have best both worlds (latency accuracy), we propose CacheNet, a model caching...
In this work, we propose to employ information-geometric tools optimize a graph neural network architecture such as the convolutional networks. More specifically, develop optimization algorithms for graph-based semi-supervised learning by employing natural gradient information in process. This allows us efficiently exploit geometry of underlying statistical model or parameter space and inference. To best our knowledge, is first work that has utilized networks can be extended other problems....
The use of a large number chemical acaricides to control these pest mites has led an increasing problem pesticide resistance, which always been the difficulty in integrated management (IPM). Fluazinam good effect on Panonychus citri, serious citrus; however, we only know mechanism action fluazinam as fungicide and its remains unclear. Through analysis using Illumina high-throughput transcriptomic sequencing differential expression genes P. citri treated with fluazinam, 59 cytochrome P450...
The Geomagnetic Induced Currents (GICs) generated in power grid by large-scale solar events may lead to electrical equipment damage and widespread outages. Consequently, it is significant achieve effective GIC prediction risk assessment for local grids. Hilbert-Huang Transform (HHT) used analyze geomagnetic components GICs monitoring data. By extracting the responses storms, instantaneous amplitudes are synthesized as equivalent disturbance indices. A combined model based on deep learning...
Sociosemiotics takes all the signs in human society into consideration including linguistic and social-cultural signs. There is still some limitation semiotic theory of Morris. The semiotics urges a new perspective to think over relationship between language social culture. Under this circumstance, sociosemiotics was developed on basis theory. This paper tries apply approach translation. It found study that sociosemiotic Translating thus regarded as message transition two systems....
We propose an extrinsic Bayesian optimization (eBO) framework for general problems on manifolds. algorithms build a surrogate of the objective function by employing Gaussian processes and utilizing uncertainty in that deriving acquisition function. This represents probability improvement based kernel process, which guides search process. The critical challenge designing manifolds lies difficulty constructing valid covariance kernels Our approach is to employ first embedding manifold onto...
In this work, we propose to train a graph neural network via resampling from graphon estimate obtained the underlying data. More specifically, or link probability matrix of is first which new will be resampled and used during training process at each layer. Due uncertainty induced resampling, it helps mitigate well-known issue over-smoothing in (GNN) model. Our framework general, computationally efficient, conceptually simple. Another appealing feature our method that requires minimal...
In this work, we propose to train a graph neural network via resampling from graphon estimate obtained the underlying data. More specifically, or link probability matrix of is first which new will be resampled and used during training process at each layer. Due uncertainty induced resampling, it helps mitigate well-known issue over-smoothing in (GNN) model. Our framework general, computationally efficient, conceptually simple. Another appealing feature our method that requires minimal...