A5100623100- Advanced Neural Network Applications
- Multimodal Machine Learning Applications
- Domain Adaptation and Few-Shot Learning
- Advanced Graph Neural Networks
- Parallel Computing and Optimization Techniques
- Advanced Image and Video Retrieval Techniques
- Digital Media Forensic Detection
- Adversarial Robustness in Machine Learning
- Neural Networks Stability and Synchronization
- Machine Learning in Materials Science
- Generative Adversarial Networks and Image Synthesis
- Topic Modeling
- Tensor decomposition and applications
- Advanced Memory and Neural Computing
- Cloud Computing and Resource Management
- Stability and Control of Uncertain Systems
- Advanced Steganography and Watermarking Techniques
- Graph Theory and Algorithms
- Video Surveillance and Tracking Methods
- Face recognition and analysis
- Music and Audio Processing
- Metal and Thin Film Mechanics
- Stochastic Gradient Optimization Techniques
- Stock Market Forecasting Methods
- Distributed and Parallel Computing Systems
Shanghai Jiao Tong University
2020-2025
Dalian University of Technology
2023-2025
China University of Mining and Technology
2022-2024
Qufu Normal University
2021-2024
Xi'an Polytechnic University
2024
ShangHai JiAi Genetics & IVF Institute
2022-2024
Chengdu University of Technology
2024
Westlake University
2021-2023
Wuhan University
2023
Renmin Hospital of Wuhan University
2023
Recent studies have shown great promise in applying graph neural networks for multivariate time series forecasting, where the interactions of are described as a structure and variables represented nodes. Along this line, existing methods usually assume that (or adjacency matrix), which determines aggregation manner network, is fixed either by definition or self-learning. However, can be dynamic evolutionary real-world scenarios. Furthermore, quite different if they observed at scales. To...
Abstract Autonomic imbalance is an important characteristic of patients after myocardial infarction (MI) and adversely contributes to post‐MI cardiac remodeling ventricular arrhythmias (VAs). A previous study proved that optogenetic modulation could precisely inhibit sympathetic hyperactivity prevent acute ischemia‐induced VAs. Here, a wireless self‐powered system introduced, which achieves long‐term precise neuromodulation in ambulatory canines. The optical based on triboelectric...
Deep learning (DL) models have achieved great success in many application domains. As such, industrial companies such as Google and Facebook acknowledged the importance of multi-tenant DL services. Although service has been studied conventional workloads, it is not deeply on deep service, especially general-purpose hardware.
Quantization is a technique to reduce the computation and memory cost of DNN models, which are getting increasingly large. Existing quantization solutions use fixed-point integer or floating-point types, have limited benefits, as both require more bits maintain accuracy original models. On other hand, variable-length uses low-bit for normal values high-precision fraction outlier values. Even though this line work brings algorithmic it also introduces significant hardware overheads due...
Large-scale vision-language pre-trained (VLP) models are prone to hallucinate non-existent visual objects when generating text based on information. In this paper, we systematically study the object hallucination problem from three aspects. First, examine recent state-of-the-art VLP models, showing that they still frequently and achieving better scores standard metrics (e.g., CIDEr) could be more unfaithful. Second, investigate how different types of image encoding in influence...
Large-scale deep neural networks (DNNs), such as large language models (LLMs), have revolutionized the artificial intelligence (AI) field and become increasingly popular. However, training or fine-tuning requires substantial computational power resources, where memory capacity of a single acceleration device like GPU is one most important bottlenecks. Owing to prohibitively overhead (e.g., 10×) GPUs' native allocator, DNN frameworks PyTorch TensorFlow adopt caching allocator that maintains...
Approximate nearest neighbor (ANN) search is a widely applied technique in modern intelligent applications, such as recommendation systems and vector databases. Therefore, efficient high-throughput execution of ANN has become increasingly important. In this paper, we first characterize the state-of-the-art product quantization-based method identify significant source inefficiency form unnecessary pairwise distance calculations accumulations. To improve efficiency, propose Juno, an end-to-end...
Abstract To achieve efficient size tuning of printed microstructures on insulating substrates, an integrated process parameter intelligent optimization design framework for alternating current pulse modulation electrohydrodynamic (AC‐EHD) printing is proposed the first time. The comprised two stages: construction a prediction model and acquisition parameters. stage employs elk herd optimizer(EHO)‐artificial neural network(ANN) to establish mapping relationship between parameters deposited...
Graph edge perturbations are dedicated to damaging the prediction of graph neural networks by modifying structure. Previous gray-box attackers employ gradients from surrogate model locate vulnerable edges perturb However, unreliability exists in on structures, which is rarely studied previous works. In this paper, we discuss and analyze errors caused structural gradients. These arise rough gradient usage due discreteness structure meta-gradient order address these problems, propose a novel...
We consider the problem of constructing linear MDS error-correcting codes with generator matrices that are sparsest and balanced. In this context, means every row has least possible number non-zero entries, balanced column contains same entries. Codes structure minimize maximal computation time computing any code symbol, a property is appealing to systems where computational load-balancing critical. The was studied before by Dau et al. it shown there always exists an over sufficiently large...
Graph neural networks (GNNs) have recently achieved remarkable success on a variety of graph-related tasks, while such relies heavily given graph structure that may not always be available in real-world applications. To address this problem, learning (GSL) is emerging as promising research topic where task-specific and GNN parameters are jointly learned an end-to-end unified framework. Despite their great progress, existing approaches mostly focus the design similarity metrics or...
Audio forgery is a critical part of digital forgery. There large variety audio forms, and copy-move one the most common but effective one. Merely word's can turn whole sentence into distinct meaning. So way to detect this form precisely effectively especially useful significant in field forensics. In paper, we propose fast method At first, segmented syllables. Then discrete Fourier transform (DFT) on each segment applied for sake extracting feature. addition, every sorted based features....
Abstract This article addresses input‐to‐state stability (ISS) for general discrete‐time impulsive and switched delayed systems coupled with impulses. In view of the multiple Lyapunov functionals subsequence method, some novel sufficient criteria pledging ISS are established. It is indicated that, as an integral will be if destabilizing impulses do not happen too frequently when system flows impulse dynamics governing delay‐dependent being ISS. Oppositely, ISS, demanded to intensively enough...
Zhuolin Yang, Wei Ping, Zihan Liu, Vijay Korthikanti, Weili Nie, De-An Huang, Linxi Fan, Zhiding Yu, Shiyi Lan, Bo Li, Mohammad Shoeybi, Ming-Yu Yuke Zhu, Bryan Catanzaro, Chaowei Xiao, Anima Anandkumar. Findings of the Association for Computational Linguistics: EMNLP 2023.
It has become cognitive inertia to employ cross-entropy loss function in classification related tasks. In the untargeted attacks on graph structure, gradients derived from attack objective are attacker's basis for evaluating a perturbation scheme. Previous methods use negative as attacking node-level models. However, suitability of constructing yet been discussed previous works. This paper argues about unreasonable perspective budget allocation. We demonstrate theoretically and empirically...
Gray-box graph attacks aim to disrupt the victim model's performance by using inconspicuous with limited knowledge of model. The details model and labels test nodes are invisible attacker. attacker constructs an imaginary surrogate trained under supervision obtain gradient on node attributes or structure. However, there is a lack discussion training models reliability provided information. general classification lose topology graph, which is, in fact, exploitable prior for This paper...