检索结果-内蒙古大学图书馆

Temporal Knowledge Question Answering via Abstract Reasoning Induction

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Chen, Ziyang Li, Dongfang Zhao, Xiang Hu, Baotian Zhang, Min Laboratory for Big Data and Decision National University of Defense Technology China Shenzhen China

In this study, we address the challenge of enhancing temporal knowledge reasoning in Large Language Models (LLMs). LLMs often struggle with this task, leading to the generation of inaccurate or misleading responses. This issue mainly arises from their limited ability to handle evolving factual knowledge and complex temporal logic. To overcome these limitations, we propose Abstract Reasoning Induction (ARI) framework, which divides temporal reasoning into two distinct phases: Knowledge-agnostic and Knowledge-based. This framework offers factual knowledge support to LLMs while minimizing the incorporation of extraneous noisy data. Concurrently, informed by the principles of constructivism, ARI provides LLMs the capability to engage in proactive, self-directed learning from both correct and incorrect historical reasoning samples. By teaching LLMs to actively construct knowledge and methods, it can significantly boosting their temporal reasoning abilities. Our approach achieves remarkable improvements, with relative gains of 29.7% and 9.27% on two temporal QA datasets, underscoring its efficacy in advancing temporal reasoning in LLMs. The code can be found at https://***/czy1999/ARI-QA. Copyright © 2023, The Authors. All rights reserved.

关键词： Knowledge based systems

Unlearning through Knowledge Overwriting: Reversible Federated Unlearning via Selective Sparse Adapter

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Zhong, Zhengyi Bao, Weidong Wang, Ji Zhang, Shuai Zhou, Jingxuan Lyu, Lingjuan Lim, Wei Yang Bryan Laboratory for Big Data and Decision National University of Defense Technology China Sony AI Japan Nanyang Technological University Singapore

Federated Learning is a promising paradigm for privacy-preserving collaborative model training. In practice, it is essential not only to continuously train the model to acquire new knowledge but also to guarantee old knowledge the right to be forgotten (i.e., federated unlearning), especially for privacy-sensitive information or harmful knowledge. However, current federated unlearning methods face several challenges, including indiscriminate unlearning of cross-client knowledge, irreversibility of unlearning, and significant unlearning costs. To this end, we propose a method named FUSED, which first identifies critical layers by analyzing each layer’s sensitivity to knowledge and constructs sparse unlearning adapters for sensitive ones. Then, the adapters are trained without altering the original parameters, overwriting the unlearning knowledge with the remaining knowledge. This knowledge overwriting process enables FUSED to mitigate the effects of indiscriminate unlearning. Moreover, the introduction of independent adapters makes unlearning reversible and significantly reduces the unlearning costs. Finally, extensive experiments on three datasets across various unlearning scenarios demonstrate that FUSED’s effectiveness is comparable to Retraining, surpassing all other baselines while greatly reducing unlearning costs. Copyright © 2025, The Authors. All rights reserved.

关键词： Federated learning

LLM-PYSC2: STARCRAFT II LEARNING ENVIRONMENT FOR LARGE LANGUAGE MODELS

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Li, Zongyuan Ni, Yanan Qi, Runnan Jiang, Lumin Lu, Chang Xu, Xiaojie Liu, Xiangbei Li, Pengfei Guo, Yunzheng Ma, Zhe Guo, Xian Huang, Kuihua Zhang, Xuebo College of Artificial Intelligence Nankai University China Laboratory for Big Data and Decision National University of Defense Technology China

This paper introduces a new environment LLM-PySC2 (the Large Language Model StarCraft II Learning Environment), a platform derived from DeepMind's StarCraft II Learning Environment that serves to develop Large Language Models (LLMs) based decision-making methodologies. This environment is the first to offer the complete StarCraft II action space, multi-modal observation interfaces, and a structured game knowledge database, which are seamlessly connected with various LLMs to facilitate the research of LLMs-based decision-making. To further support multi-agent research, we developed an LLM collaborative framework that supports multi-agent concurrent queries and multi-agent communication. In our experiments, the LLM-PySC2 environment is adapted to be compatible with the StarCraft Multi-Agent Challenge (SMAC) task group and provided eight new scenarios focused on macro-decision abilities. We evaluated nine mainstream LLMs in the experiments, and results show that sufficient parameters are necessary for LLMs to make decisions, but improving reasoning ability does not directly lead to better decision-making outcomes. Our findings further indicate the importance of enabling large models to learn autonomously in the deployment environment through parameter training or train-free learning techniques. Ultimately, we expect that the LLM-PySC2 environment can promote research on learning methods for LLMs, helping LLM-based methods better adapt to task scenarios. Copyright © 2024, The Authors. All rights reserved.

关键词： Multi agent systems

A method for accelerating relations search over big data scenario 5

学校读者我要写书评

暂无评论

A method for accelerating relations search over big data sce...

5th International Conference on Algorithms, Computing and Artificial Intelligence, ACAI 2022

作者： Liu, Lihua Wang, Mao Xiao, Kaiming Chen, Haiwen College of System Engineering National University of Defense Technology China and Laboratory for Big Data and Decision National University of Defense Technology Changsha410073 China

ISBN: (纸本)9781450398343

We propose an acceleration method for relations search over big data scenario. The relations search is in essence an induced subgraph search of graph data model, which is to search all edges with all adjacent vertices falling into a given vertex set. For example, in the risk-control scenario for anti-fraud, for a swindling gang of highly similar fraud behaviors, figuring out all relations between these swindlers can be greatly beneficial to precisely attack all similar gangs. In-memory induced subgraph search requires lots of random accesses, while in big data scenario, relations data tend to be stored on disk when the memory is limited, and the corresponding random lookups over disk incorporate huge overhead. Also, even a singleton disk access could be costly since there are usually many properties data in a relation. Hence, existing methods would suffer considerable performance bottlenecks when applied over data on disk. We build graph over disk-based edge index, and propose multiple BFS (Bread First Search) level gaps based induces subgraph acceleration method, with efficient performance guarantee when the memory is limited. We avoid imbalance in performance for query evaluation with edge index based data organization, with which we further reduce redundant I/O accesses. We filter most invalid no-result edge queries based our BFS level gaps framework. We find that, for each BFS, we can confirm the no edge status between a pair of vertices if their BFS accessing level are of distance more than 1, which indicates the nonexistence of edge. Extensive experiments over real world graph datasets confirm the effectiveness of our method. We find that more than 97% invalid edge queries are filtered by our method, which greatly improve the system performance. Hence, our method with multiple BFS level gap encoding over vertices could greatly filter no-result edge queries and improve the relation search performance. Out method also reduce the redundant I/O access over disk wi

关键词： big data

Extracting Events Using Spans Enhanced with Trigger-Argument Interaction 18

学校读者我要写书评

暂无评论

Extracting Events Using Spans Enhanced with Trigger-Argument...

18th International Conference on Intelligent Systems and Knowledge Engineering, ISKE 2023

作者： Xu, Chi Zeng, Zefan Duan, Jiashun Qi, Qi Zhang, Xin National University of Defense Technology Science and Technology on Information Systems Engineering Laboratory Changsha China National University of Defense Technology Big Data and Decision Laboratory Changsha China War Studies College The Academy of Military Sciences Beijing China

ISBN: (纸本)9798350318401

Event extraction (EE) aims to detect event triggers and extract their corresponding arguments. In more realistic situations, events often come in a more complicated way in which multiple events may appear simultaneously and arguments may have overlaps in a sentence. For languages without delimiters like Chinese, most existing EE models are token-level and thus may have difficulties in solving multi-event or overlapping problems. What's more, they are prone to overlook the word-level or span-level features and the rich interactions between triggers and arguments. To tackle these problems, we propose a span-based model for multiple and overlapped EE with trigger-argument interaction enhanced span representations. Experimental results show that our approach outperforms other mainstream baselines on the DuEE dataset. © 2023 IEEE.

关键词： multiple event overlapped event span-based model trigger-argument interaction

The Application of big data and Machine Learning in Petrogeochemistry

学校读者我要写书评

暂无评论

The Application of Big Data and Machine Learning in Petrogeo...

International Conference on big data and Information Analytics (bigDIA)

作者： Shengkai Jin Yixing Ma Shuqi Yang Xiaosong Li Command Center for Intergrated Natural Resources Survey China Goelogical Survey Laboratory for Big Data and Decision Beijing China

ISBN: (数字)9798350354621

ISBN: (纸本)9798350354638

With the advancement of big data and artificial intelligence technologies, geoscience research has entered a new phase. This paper examines the application of machine learning in petrogeochemistry, addressing areas such as geodynamic numerical simulation, metallogenic geological process modeling, tectonic environment discrimination of igneous, magma provenance, metallogenic prediction and ore deposit genetic type classification. Unlike traditional two-dimensional or three-dimensional methods of discrimination and visualization, machine learning techniques efficiently manage large-scale and high-dimensional data, greatly enhancing both accuracy and efficiency. This paper highlights the benefits of machine learning in petrological geochemistry research and provides recommendations for improving data processing and model application, supported by case studies. Ultimately, it asserts that the integration of machine learning with big data offers geologists powerful tools, facilitating the advanced development of petrological geochemistry research.

关键词： Accuracy Mineralization Machine learning big data Predictive models Numerical simulation Numerical models Geodynamics Tectonics Magma

Predicting Critically Ill Patients Short-Term Mortality Risk Using Routinely Collected data:Deep Learning Model Development,Validation,and Explanation

学校读者我要写书评

暂无评论

Journal of Systems Science and Information 2023年第3期11卷 365-377页

作者： Shangping ZHAO Pan LIU Guohui LI Yanming GUO Guanxiu TANG Laboratory for Big Data and Decision National University of Defense TechnologyChangsha 410003China Department of Critical Care Medicine the Third Xiangya HospitalCentral South UniversityChangsha 410013China Nursing Department the Third Xiangya HospitalCentral South UniversityChangsha 410013China

This paper aims to develop and validate a deep learning-based short-term mortality risk prediction model for critically ill patients by using routinely collected data in a large Chinese cohort and explore the explainability of the model decision.A total of 10925 critically ill patients between January 2014 and June 2020 are included in this *** routinely collected in the electronic health records(EHRs)system are extracted and used to develop a short-term mortality risk prediction model based on a deep artificial neural network(ANN).The features include demographic characteristics,vital signs,laboratory tests,and the daily dose of intravenous *** developed deep learning model(AUROC:0.88,AUPRC:0.63,Brier score:0.108)is superior to the model based on APACHEⅡscores(AUROC:0.78,AURPC:0.52,Brier score:0.124)in the prediction of hospital mortality for critically ill *** attribution analysis based on the integrated gradients method shows that measurements observed at a later time seem to have a more significant influence on mortality,while earlier usage of amiodarone or dexmedetomidine contributed to lower *** well-performing and interpretable model may have practical implications for improving the quality of care for critically ill patients.

关键词： Critically ill patients deep learning artificial neural network mortality prediction APACHEⅡ

Accumulative Time Based Ranking Method to Reputation Evaluation in Information Networks

学校读者我要写书评

暂无评论

Journal of Computer Science & Technology 2022年第4期37卷 960-974页

作者： Hao Liao Qi-Xin Liu Ze-Cheng Huang Ke-Zhong Lu Chi Ho Yeung Yi-Cheng Zhang National Engineering Laboratory on Big Data System Computing Technology College of Computer Science and SoftwareEngineeringShenzhen UniversityShenzhen 518060China Guangdong Province Key Laboratory of Popular High Performance Computers Shenzhen UniversityShenzhen 518060China Guangdong Province Engineering Center of China-Made High Performance Data Computing System College of ComputerScience and Software EngineeringShenzhen UniversityShenzhen 518060China Institute of Big Data Intelligent Management and Decision Shenzhen UniversityShenzhen 518060China Department of Science and Environmental Studies The Education University of Hong KongHong Kong 999077China Department of Physics University of FribourgFribourg 1700Switzerland

Due to over-abundant information on the Web, information filtering becomes a key task for online users to obtain relevant suggestions and how to extract the most related item is always a key topic for researchers in various fields. In this paper, we adopt tools used to analyze complex networks to evaluate user reputation and item quality. In our proposed Accumulative Time Based Ranking (ATR) algorithm, we take into account the growth record of the network to identify the evolution of the reputation of users and the quality of items, by incorporating two behavior weighting factors which can capture the hidden facts on reputation and quality dynamics for each user and item respectively. Our proposed ATR algorithm mainly combines the iterative approach to rank user reputation and item quality with temporal dependence compared with other reputation evaluation methods. We show that our algorithm outperforms other benchmark ranking algorithms in terms of precision and robustness on empirical datasets from various online retailers and the citation datasets among research publications. Therefore, our proposed method has the capability to effectively evaluate user reputation and item quality.

关键词： temporal network behavior dynamics reputation evaluation ranking algorithm

Hierarchical Expert Prompt for Large-Language-Model: An Approach Defeat Elite AI in TextStarCraft II for the First Time

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Li, Zongyuan Lu, Chang Xu, Xiaojie Qi, Runnan Ni, Yanan Jiang, Lumin Liu, Xiangbei Zhang, Xuebo Fang, Yongchun Huang, Kuihua Guo, Xian College of Artificial Intelligence Nankai University Tianjing China Laboratory for Big Data and Decision National University of Defense Technology Changsha China

Since the emergence of the Large Language Model (LLM), LLM has been widely used in fields such as writing, translating, and searching. However, there is still great potential for LLM-based methods in handling complex tasks such as decision-making in the StarCraft II environment. To address problems such as lack of relevant knowledge and poor control over subtasks of varying importance, we propose a Hierarchical Expert Prompt (HEP) for LLM. Our method improves the understanding of game situations through expert-level tactical knowledge, improving the processing quality of tasks of varying importance through a hierarchical framework. Our approach defeated the highest level (Elite) standard built-in agent in TextStarCraft II for the first time and consistently outperformed the baseline method[1] in other difficulties. Our experiments suggest that the proposed method is a practical solution for tackling complex decision-making challenges. The replay video can be viewed on https://***/video/BV1uz42187EF and https://***/dO3PshWLV5M, and our codes have been open-sourced on https://***/luchang1113/HEP-LLM-play-StarCraftII. Copyright © 2025, The Authors. All rights reserved.

关键词： Modeling languages