A5087059403- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Natural Language Processing Techniques
- Multimodal Machine Learning Applications
- Manufacturing Process and Optimization
- Topic Modeling
- Speech Recognition and Synthesis
- Educational Technology and Assessment
- Advanced Numerical Analysis Techniques
- Handwritten Text Recognition Techniques
- Advanced Manufacturing and Logistics Optimization
- Visual Attention and Saliency Detection
- COVID-19 diagnosis using AI
Sun Yat-sen University
2020-2022
An intelligent toolpath (processing sequence) planning for fabricating aircraft structural parts is developed to obtain a minimum-length while maintaining the required machining time. The problem first solved in following ways. Firstly, an abstraction of process proposed pick up feed/retract points each feature be machined as calculation basis. Secondly, chaotic simulated annealing method improved through storing optimal solution, adding control condition and selecting parameter reasonably...
Few-shot object detection is an important but challenging task where only a few instances of novel categories are available. The widely used approach to pretrain detector on base classes with abundant samples and then fine-tune it for classes. Due the extreme data imbalance between classes, performance degrades distribution bias. To overcome this limitation, we propose calibration strategy class discrimination regularization method better few-shot detection. Based theoretical analysis...
Extending the context length (i.e., maximum supported sequence length) of LLMs is paramount significance. To facilitate long training LLMs, parallelism has emerged as an essential technique, which scatters each input across multiple devices and necessitates communication to process sequence. In essence, existing methods assume homogeneous lengths all sequences are equal in therefore leverages a single, static scattering strategy for sequences. However, reality, LLM corpora exhibit...
Current state-of-the-art methods usually utilize feature pyramid to provide various receptive fields for detecting objects at different scales. However, the maps from low- high-level layers have large semantic gaps and are with spatial resolutions, so that their representational capacity differs noise is introduced when fusing them. To overcome this limitation carry out better object detection, we design a novel network named Receptive Field Pyramid Network (RFPN). The proposed method...
Few-shot object detection is a promising approach to solving the problem of detecting novel objects with only limited annotated data for training. Most existing methods are developed based on progress in few-shot classification, which pay little attention improving localization module and modelling class interrelation. To address these issues, this paper proposes two modules, namely Region-interactive Proposal Network (Ri-PN) Class-interactive Feature Learning (Ci-FL), better classification...