A5100384181- Maritime Transport Emissions and Efficiency
- Multimodal Machine Learning Applications
- Advanced Battery Technologies Research
- Topic Modeling
- Domain Adaptation and Few-Shot Learning
- Maritime Ports and Logistics
- Machine Learning and ELM
- Text and Document Classification Technologies
- Natural Language Processing Techniques
Wuhan University
2024
Yanshan University
2024
Nanjing University of Posts and Telecommunications
2019
The applications of data augmentation in natural language processing have been limited. In this paper, we propose a novel method named Hierarchical Data Augmentation (HDA) which applied for text classification. Firstly, inspired by the hierarchical structure texts, as words form sentence and sentences document, HDA implements strategy augmenting texts at word-level level respectively. Secondly, cropping, popular computer vision, each level, utilizes attention mechanism to distill (crop)...
Class-incremental learning (CIL) aims to avoid catastrophic forgetting without relying on task identifiers. Exemplar-free CIL (EFCIL) poses further challenges due no access historical training data. A recent addition EFCIL is self-supervised learning, promoting diverse feature representations from a broader range of classes. However, limited diversity, high computational overhead, and practical issues with large-scale augmented datasets remain challenges. Moreover, existing frameworks focus...