A5100400133- Adversarial Robustness in Machine Learning
- Natural Language Processing Techniques
- Topic Modeling
- Anomaly Detection Techniques and Applications
- Privacy-Preserving Technologies in Data
- Advanced Malware Detection Techniques
- Multimodal Machine Learning Applications
- Domain Adaptation and Few-Shot Learning
- Advanced Graph Neural Networks
- Advanced Neural Network Applications
- Privacy, Security, and Data Protection
- Network Security and Intrusion Detection
- Reinforcement Learning in Robotics
- Imbalanced Data Classification Techniques
- Financial Distress and Bankruptcy Prediction
- Speech and dialogue systems
- Cryptography and Data Security
- Human Mobility and Location-Based Analysis
- AI-based Problem Solving and Planning
- Digital Media Forensic Detection
- Security and Verification in Computing
- Text and Document Classification Technologies
- Robot Manipulation and Learning
- Data Stream Mining Techniques
- Technology Use by Older Adults
China Agricultural University
2025
Xi'an Jiaotong University
2023-2024
Data61
2020-2024
Commonwealth Scientific and Industrial Research Organisation
1999-2024
Indiana University – Purdue University Indianapolis
2024
Shanghai Jiao Tong University
2020-2024
University of Birmingham
2009-2024
University of Indianapolis
2024
Hebei University
2021-2024
Washington University in St. Louis
2024
Recently, significant accuracy improvement has been achieved for acoustic recognition systems by increasing the model size of Long Short-Term Memory (LSTM) networks. Unfortunately, ever-increasing LSTM leads to inefficient designs on FPGAs due limited on-chip resources. The previous work proposes use a pruning based compression technique reduce and thus speedups inference FPGAs. However, random nature transforms dense matrices highly unstructured sparse ones, which unbalanced computation...
Transfer learning provides an effective solution for feasibly and fast customize accurate <i>Student</i> models, by transferring the learned knowledge of pre-trained <i>Teacher</i> models over large datasets via fine-tuning. Many Teacher used in transfer are publicly available maintained public platforms, increasing their vulnerability to backdoor attacks. In this article, we demonstrate a threat tasks on both image time-series data leveraging accessible aimed at defeating three commonly...
The AI City Challenge was created with two goals in mind: (1) pushing the boundaries of research and development intelligent video analysis for smarter cities use cases, (2) assessing tasks where level performance is enough to cause real-world adoption. Transportation a segment ripe such fifth attracted 305 participating teams across 38 countries, who leveraged city-scale real traffic data high-quality synthetic compete five challenge tracks. Track 1 addressed video-based automatic vehicle...
Confidence calibration, which aims to make model predictions equal the true correctness measures, is important for neural machine translation (NMT) because it able offer useful indicators of errors in generated output. While prior studies have shown that NMT models trained with label smoothing are well-calibrated on ground-truth training data, we find miscalibration still remains a severe challenge during inference due discrepancy between and inference. By carefully designing experiments...
As data continue to grow in complexity and size, there is an imperative need for more efficient robust storage solutions. DNA has emerged as a promising avenue solve this problem, but existing approaches do not perform efficiently enough on video data, particularly information density time efficiency. This paper introduces VSD, pioneering encoding method segmentation DNA, leveraging the Reed–Solomon (RS) error correction code. addresses these limitations through innovative combination of...
Deep-learning-based identity management systems, such as face authentication are vulnerable to adversarial attacks. However, existing attacks typically designed for single-task purposes, which means they tailored exploit vulnerabilities unique the individual target rather than being adaptable multiple users or systems. This limitation makes them unsuitable certain attack scenarios, morphing, universal, transferable, and counter In this paper, we propose a multi-task algorithm called MTADV...
De novo peptide sequencing directly identifies peptides from mass spectrometry data, playing a critical role in discovering novel proteins and analyzing complex biological samples without reliance on existing databases. To address challenges both speed accuracy, transformer-based model, TSARseqNovo, incorporates two key innovations: Semi-Autoregressive decoder for parallel prediction of multiple amino acids Masking Refinement refining low-confidence predictions. These features significantly...
This research introduces a novel text generation model that combines BERT's semantic interpretation strengths with GPT-4's generative capabilities, establishing high standard in generating coherent, contextually accurate language. Through the combined architecture, enhances depth and maintains smooth, human-like flow, overcoming limitations seen prior models. Experimental benchmarks reveal BERT-GPT-4 surpasses traditional models, including GPT-3, T5, BART, Transformer-XL, CTRL, key metrics...
Abstract Background In this study, we examined psychometric properties of the Chinese version Resilience Scale (RS) and parenting-related factors associated with resilience among disaster-exposed adolescents. Methods Eighteen months after earthquake, a total 1266 adolescents (43.4% male, mean age = 15.98; SD 1.28) were assessed using RS, Post-traumatic Stress Disorder Self-Rating Scale, Depression Self-rating for Children, Screen Child Anxiety Related Emotional Disorders, Parental Bonding...
Graph Neural Networks (GNNs) have emerged as a series of effective learning methods for graph-related tasks. However, GNNs are shown vulnerable to adversarial attacks, where attackers can fool into making wrong predictions on samples with well-designed perturbations. Specifically, we observe that the current evasion attacks suffer from two limitations: (1) attack strategy based reinforcement method might not be transferable when budget changes; (2) greedy mechanism in vanilla gradient-based...
The futures market's forecasts are significant to investors and policymakers, where the application of deep learning approaches finance has received a great deal attention. In this study, we propose multivariate financial time-series forecasting method. Our model addresses long- short-term features, multimodal non-stationarity nature by incorporating improved neural networks certified noise injection. Specifically, variational autoencoder is used extract high-level features time-series,...
In Federated Learning (FL), blockchain has been extensively used to achieve distributed and tamper-resistant data processing. However, typical Blockchain-based (BFL) rarely considers clients' resource computing limits. High-capacity clients may be sacrificed when all train on the same neural network. This paper proposes a Heterogeneous (BlocFL) model address challenges above. BlocFL replaces central server with consortium blockchain, several networks are employed for local training....
Deep learning models with backdoors act maliciously when triggered but seem normal otherwise. This risk, often increased by model outsourcing, challenges their secure use. Although countermeasures exist, defense against adaptive attacks is under-examined, possibly leading to security misjudgments. study the first intricate examination illustrating difficulty of detecting in outsourced models, especially attackers adjust strategies, even if capabilities are significantly limited. It...
Detection-based defense approaches are effective against adversarial attacks without compromising the structure of protected model. However, they could be bypassed by stronger and limited in their ability to handle high-fidelity images. In this paper, we explore an detection-based on images (including high-resolution images) extending investigation beyond a single-instance perspective incorporate its transformations as well. Our intuition is that essential characteristics valid image...