A5019940038- Context-Aware Activity Recognition Systems
- Anomaly Detection Techniques and Applications
- Non-Invasive Vital Sign Monitoring
- Natural Language Processing Techniques
- Evaluation and Optimization Models
- Human Pose and Action Recognition
- EEG and Brain-Computer Interfaces
- Hate Speech and Cyberbullying Detection
- Advanced Computational Techniques and Applications
- Advanced Decision-Making Techniques
- Advanced Measurement and Detection Methods
- Remote-Sensing Image Classification
- Sentiment Analysis and Opinion Mining
- Misinformation and Its Impacts
- Image and Video Stabilization
- Laser and Thermal Forming Techniques
- Constraint Satisfaction and Optimization
- IoT and Edge/Fog Computing
China Electronics Technology Group Corporation
2024
Nanjing Normal University
2023-2024
During recent years, deep convolutional neural networks have demonstrated dominant performance in human activity recognition (HAR) using wearable sensors. However, they often come at high computational cost when fueled with fixed-length sliding window. This article primarily aims to accelerate inference from a novel perspective of reducing temporal redundancy sensor data. Inspired by the fact that not all time intervals within window are activity-relevant, we formulate prediction problem as...
Hateful memes have become a significant concern on the Internet, necessitating robust automated detection systems. While large multimodal models shown strong generalization across various tasks, they exhibit poor to hateful meme due dynamic nature of tied emerging social trends and breaking news. Recent work further highlights limitations conventional supervised fine-tuning for in this context. To address these challenges, we propose Large Multimodal Model Retrieval-Guided Contrastive...
During the past decade, deep convolutional neural networks (CNNs) have shown remarkable success in ubiquitous human activity recognition (HAR) scenarios. However, prior most works are static, which to manually predefine a fixed kernel size before training, hence requiring time-consuming intervention. A more efficient solution is determine during training stage, rather than design stage. Different from previous static designs, this article proposes an adaptive called dynamic Gaussian...
We derive and investigate two DPO variants that explicitly model the possibility of declaring a tie in pair-wise comparisons. replace Bradley-Terry with well-known modeling extensions, by Rao Kupper Davidson, assign probability to ties as alternatives clear preferences. Our experiments neural machine translation summarization show labeled can be added datasets for these without degradation task performance is observed when same tied pairs are presented DPO. find empirically inclusion leads...
Recent years have witnessed great success of deep convolutional networks in sensor-based human activity recognition (HAR), yet their practical deployment remains a challenge due to the varying computational budgets required obtain reliable prediction. This paper focuses on adaptive inference from novel perspective signal frequency, which is motivated by an intuition that low-frequency features are enough for recognizing “easy” samples, while only “hard” samples need temporally detailed...
Minimum Bayes Risk (MBR) decoding can significantly improve translation performance of Multilingual Large Language Models (MLLMs). However, MBR is computationally expensive. We show how the recently developed Reinforcement Learning technique, Direct Preference Optimization (DPO), fine-tune MLLMs to get gains without any additional computation in inference. Our method uses only a small monolingual fine-tuning set and yields improved on multiple NMT test sets compared DPO.