A5106762821- Topic Modeling
- Natural Language Processing Techniques
- Quantum Information and Cryptography
- AI in cancer detection
- Domain Adaptation and Few-Shot Learning
- Quantum Computing Algorithms and Architecture
- Multimodal Machine Learning Applications
- Quantum Mechanics and Applications
- Anomaly Detection Techniques and Applications
- Time Series Analysis and Forecasting
- Machine Learning and ELM
- Educational Technology and Assessment
- Higher Education and Teaching Methods
- Direction-of-Arrival Estimation Techniques
- Advanced Computational Techniques and Applications
- Cervical Cancer and HPV Research
- Target Tracking and Data Fusion in Sensor Networks
- Software System Performance and Reliability
- Technology and Security Systems
- Reinforcement Learning in Robotics
- Second Language Learning and Teaching
- Machine Learning and Algorithms
- COVID-19 diagnosis using AI
- Neural Networks and Applications
- Advanced Neural Network Applications
Microsoft Research (United Kingdom)
2024-2025
China University of Geosciences
2015-2024
Central South University
2007-2024
Wuyi University
2023-2024
Microsoft (Finland)
2022-2023
Henan University of Science and Technology
2022
PLA Information Engineering University
2021
Institute of Computing Technology
2020
Chinese Academy of Sciences
2020
University of Chinese Academy of Sciences
2020
Quantum entanglement is a fundamental physical resource that enables range of information-processing tasks with no efficient solution on classical computers. In this paper, we construct class general maximum entangled states by parameterizing the coefficients Bell states. By calculating, obtain local unitary operations required to realize mutual transformation constructed states, analogous Pauli operators acting one particle state. Interestingly, two particles are identical, whereas for they...
Recent research has demonstrated that the multi-task fine-tuning of multi-modal Large Language Models (LLMs) using an assortment annotated downstream vision-language datasets significantly enhances their performance. Yet, during this process, a side effect, which we termed as "multi-modal alignment tax", surfaces. This effect negatively impacts model's ability to format responses appropriately - for instance, its "politeness" due overly succinct and unformatted nature raw annotations,...
Document retrieval techniques form the foundation for development of large-scale information systems. The prevailing methodology is to construct a bi-encoder and compute semantic similarity. However, such scalar similarity difficult reflect enough impedes our comprehension results. In addition, this computational process mainly emphasizes global semantics ignores fine-grained relationship between query complex text in document. paper, we propose new method called Generation Augmented...
Abstract Controlled bidirectional quantum secure direct communication (CBQSDC) is an important research direction in communication. It can enable two legitimate users to confidentially exchange secret messages only with the permission of controller. In many practical scenarios, CBQSDC needs avoid controller from obtaining useful information on messages. Therefore, this paper, we introduce concept independent controlled (CICBQSDC) that has a control but restrict obtain other words, require...
Parameter-efficient fine-tuning (PEFT) has been widely employed for domain adaptation, with LoRA being one of the most prominent methods due to its simplicity and effectiveness. However, in multi-task learning (MTL) scenarios, tends obscure distinction between tasks by projecting sparse high-dimensional features from different into same dense low-dimensional intrinsic space. This leads task interference suboptimal performance variants. To tackle this challenge, we propose MTL-LoRA, which...
Cervical abnormal cell detection is a challenging task as the morphological discrepancies between and normal cells are usually subtle. To determine whether cervical or abnormal, cytopathologists always take surrounding references to identify its abnormality. mimic these behaviors, we propose explore contextual relationships boost performance of detection. Specifically, both cell-to-global images exploited enhance features each region interest (RoI) proposal. Accordingly, two modules, dubbed...
Daixuan Cheng, Shaohan Huang, Junyu Bi, Yuefeng Zhan, Jianfeng Liu, Yujing Wang, Hao Sun, Furu Wei, Weiwei Deng, Qi Zhang. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.
In the process of product design, engineers usually find it is difficult to precisely and reuse others' empirical knowledge resources, especially lesson-learned knowledge, which not well collected by organisation. This study proposes a novel approach, uses semantic-based visualised wiki system (SVWkS) support reuse. The core search framework topic map. architecture this map creation method designed, has five major modules: items pre-processing, extraction, relation computation, weight...
Deep learning technology provides novel solutions for increasingly complex target tracking requirements. For traditional models, the movement of need to be simulated by a predefined mathematical model. However, it is extremely difficult obtain sufficient information in advance, which makes challenging track changeable and noisy trajectories timely precise manner. A deep framework constructed automatic trajectory based on dynamic laws motion, called DeepGTT. Specifically, generator mapper...
Nuclei segmentation is an essential step in DNA ploidy analysis by image-based cytometry (DNA-ICM) which widely used cytopathology and allows objective measurement of content (ploidy). The routine fully supervised learning-based method requires often tedious expensive pixel-wise labels. In this paper, we propose a novel weakly nuclei framework exploits only sparsely annotated bounding boxes, without any key to integrate the traditional image self-training into instance segmentation. We first...
Zhijiang Baijiu is a Chinese strong flavour liquor produced using traditional solid state fermentation involving microorganisms in pit mud on cellar walls. In this study, samples were collected from three different cellars distillery ranging age 10 to 20 and 30 years. The bacterial communities analysed denaturing gradient gel electrophoresis, high-throughput sequencing quantitative real time PCR. These analyses showed that the diversity of community was relatively stable aged years old. High...
Abstract From the perspective of resource theory, it is interesting to achieve same quantum task using as few resources possible. Semiquantum key distribution (SQKD), which allows a user share confidential with classical who prepares and operates qubits on only one basis, an important example for studying this issue. To further limit used by users, in paper, first SQKD protocol constructed, restricts prepare states basis removes user's measurement capability. Furthermore, proven that...
Parameter-efficient fine-tuning (PEFT) has been widely employed for domain adaptation, with LoRA being one of the most prominent methods due to its simplicity and effectiveness. However, in multi-task learning (MTL) scenarios, tends obscure distinction between tasks by projecting sparse high-dimensional features from different into same dense low-dimensional intrinsic space. This leads task interference suboptimal performance variants. To tackle this challenge, we propose MTL-LoRA, which...
In-context learning (ICL) allows large language models (LLMs) to adapt new tasks directly from the given demonstrations without requiring gradient updates. While recent advances have expanded context windows accommodate more demonstrations, this approach increases inference costs necessarily improving performance. To mitigate these issues, We propose StreamAdapter, a novel that updates model parameters at test time, eliminating need for explicit in-context demonstrations. StreamAdapter...
In-context learning (ICL) allows large language models (LLMs) to adapt new tasks directly from the given demonstrations without requiring gradient updates. While recent advances have expanded context windows accommodate more demonstrations, this approach increases inference costs necessarily improving performance. To mitigate these issues, We propose StreamAdapter, a novel that updates model parameters at test time, eliminating need for explicit in-context demonstrations. StreamAdapter...
Abstract The computation of traveltimes for compressional (P) waves is significantly important earthquake and seismic exploration applications. eikonal equation, functioning as a fundamental tool in wave propagation analysis, plays crucial role the construction images. However, inherent nonlinearity equation introduces complexity, particularly when addressing anisotropic media. derived directly from Christoffel applicable general traveltime calculations. its complex quartic presenting...