检索结果-内蒙古大学图书馆

arXiv 2024年

作者： Yang, Haotong Hu, Yi Kang, Shijia Lin, Zhouchen Zhang, Muhan School of Intelligence Science and Technology Peking University China Institution for Artificial Intelligence Peking University China State Key Lab of General Artificial Intelligence Peking University China Guangdong Guangzhou China

Large language models (LLMs) can solve an increasing number of complex reasoning tasks while making surprising mistakes in basic numerical understanding and processing (such as 9.11 > 9.9). The latter ability is essential for tackling complex arithmetic and mathematical problems and serves as a foundation for most reasoning tasks, but previous work paid little attention to it or only discussed several restricted tasks (like integer addition). In this paper, we comprehensively investigate the numerical understanding and processing ability (NUPA) of LLMs. Firstly, we introduce a benchmark covering four common numerical representations and 17 distinct numerical tasks in four major categories, resulting in 41 meaningful combinations in total. These tasks are derived from primary and secondary education curricula, encompassing nearly all everyday numerical understanding and processing scenarios, and the rules of these tasks are very simple and clear. Through the benchmark, we find that current LLMs fail frequently in many of the tasks. To study the problem, we train small models with existing and potential techniques for enhancing NUPA (such as tokenizers, positional encoding, and number formats), comprehensively evaluating their effectiveness using our testbed. We also finetune practical-scale LLMs on our proposed NUPA tasks and find that 1) naive finetuning can significantly improve NUPA on many but not all tasks, and 2) surprisingly, techniques designed to enhance NUPA prove ineffective for finetuning pretrained models. We further explore the impact of chain-of-thought techniques on NUPA. Our work provides a more detailed and comprehensive understanding of NUPA in LLMs. Our benchmark and codes are released at https://***/GraphPKU/number_cookbook. Copyright © 2024, The Authors. All rights reserved.

关键词： Benchmarking

来源：评论

学校读者我要写书评

暂无评论

U-DiTs: Downsample Tokens in U-Shaped Diffusion Transformers

arXiv

引用

arXiv 2024年

作者： Tian, Yuchuan Tu, Zhijun Chen, Hanting Hu, Jie Xu, Chao Wang, Yunhe State Key Lab of General AI School of Intelligence Science and Technology Peking University China Huawei Noah’s Ark Lab Hong Kong

Diffusion Transformers (DiTs) introduce the transformer architecture to diffusion tasks for latent-space image generation. With an isotropic architecture that chains a series of transformer blocks, DiTs demonstrate competitive performance and good scalability;but meanwhile, the abandonment of U-Net by DiTs and their following improvements is worth rethinking. To this end, we conduct a simple toy experiment by comparing a U-Net architectured DiT with an isotropic one. It turns out that the U-Net architecture only gain a slight advantage amid the U-Net inductive bias, indicating potential redundancies within the U-Net-style DiT. Inspired by the discovery that U-Net backbone features are low-frequency-dominated, we perform token downsampling on the query-key-value tuple for self-attention that bring further improvements despite a considerable amount of reduction in computation. Based on self-attention with downsampled tokens, we propose a series of U-shaped DiTs (U-DiTs) in the paper and conduct extensive experiments to demonstrate the extraordinary performance of U-DiT models. The proposed U-DiT could outperform DiT-XL/2 with only 1/6 of its computation cost. Codes are available at https://***/YuchuanTian/U-DiT. Copyright © 2024, The Authors. All rights reserved.

关键词： Diffusion

来源：评论

学校读者我要写书评

暂无评论

AWNet: Negative obstacle semantic segmentation Network based on adaptive weight loss 22

AWNet: Negative obstacle semantic segmentation Network based...

引用

22nd IEEE International Conference on Industrial Informatics, INDIN 2024

作者： Zhao, Jiarui Yu, Guizhen Wang, Jie Wang, Zhangyu Zhang, Junjie Feng, Yunsong School of Transportation Science and Engineering Beihang University Beijing China State Key Lab of Intelligent Transportation System Beihang University Beijing China The Key Laboratory of Autonomous Transportation Technology for Special Vehicles Beihang University Beijing China Ministry of Industry and Information Technology Beihang University Beijing China Research Institute for Frontier Science Beihang University Hefei China Hefei innovation research institute of Beihang university Beihang University Hefei China State Key Laboratory of Pulsed Power Laser Technology National University of Defense Technology Heufei China

ISBN: (纸本)9798331527471

On roads, both vehicles and pedestrians will leave traces after passing by. Slight traces will not cause any trouble to the passing vehicles and pedestrians subsequently. However, there are certain potholes in the road that have been formed by repeated crushing of vehicles. It may cause mechanical damage to following vehicles and lead to accidents as vehicles get stuck in the potholes. Thus, the segmentation of negative obstacles is crucial for the safe driving of self-driving vehicles. This paper proposed an AWNet neural network to this end, and quantitative analyses were conducted based on the publicly available dataset DRNO. Specifically, a random cut-and-joint module (RCJ) was proposed to augment the dataset. The cross-entropy loss function was modified by adding an adaptive weight module, i.e., the adaptive weight cross-entropy loss function. It allows the model to focus on negative obstacles during implementation. Furthermore, this paper proposed a post-processing module for adaptive viewpoint conversion based on depth maps, which is capable of fusing depth maps and RGB images to convert the front view into a BEV viewpoint. The mAcc for Negative obstacle semantic segmentation can reach 78.39%, the mIoU can reach 73.21%, the mF1 can reach 79.23%. The experimental results reveal that the AWNet is able to state-of-the-art performance compared with other networks. © 2024 IEEE.

关键词： Semantic Segmentation

来源：评论

学校读者我要写书评

暂无评论

Impact Analysis of Passive Multi- False Targets to Phased Array Radar Resources 15

Impact Analysis of Passive Multi- False Targets to Phased Ar...

引用

15th International Conference on Microwave and Millimeter Wave technology, ICMMT 2023

作者： Hao, Guo-Qing Feng, De-Jun Cheng, Bi-Qin Wang, Jun-Jie Wang, Ling National University of Defense Technology The State Key Lab. of Complex Electromagnetic Environ. Effects on Electronics and Information System College of Electronic Science Changsha410073 China Unit 95861 of Pla Jiuquan735000 China

ISBN: (纸本)9798350338874

The phase-switched screen (PSS) can change the reflected echo of the target through a certain control method, thereby generating false targets. The study of material properties and the impact of materials on radar has attracted a lot of attention from the academic community. However, there are few studies on the influence of signal echo changes caused by artificial surface metamaterials on radar resources. In this paper, using PSS to change radar echo information to generate false targets in the Doppler frequency domain. Then the multi-false targets jamming effect on radar resources and multi-target tracking is analyzed. Simulation results show that these targets can consume radar search and tracking resources, generate a large number of temporary trails. This research is beneficial to the application of PSS for interference radar resources. © 2023 IEEE.

关键词： Target tracking

来源：评论

学校读者我要写书评

暂无评论

Dense Voxel Representation Network for Implicit Scene Completion

Dense Voxel Representation Network for Implicit Scene Comple...

引用

IEEE International Conference on Multimedia and Expo (ICME)

作者： Fan Dai Yun Zhu Yaqi Shen Jin Xie Jianjun Qian PCA Lab Nanjing University of Science and Technology Nanjing China State Key Laboratory for Novel Software Technology & School of Intelligence Science and Technology Nanjing University China

ISBN: (数字)9798350390155

ISBN: (纸本)9798350390162

Implicit scene completion aims to learn an implicit representation of dense point clouds from incomplete ones. Since point clouds are disordered and irregular, some implicit scene completion methods learn representations from voxelized point clouds with sparse convolution. Despite achieving promising results, they lack deep exploration of feature learning on empty voxels, which is beneficial for implicit scene completion task. To address this, we propose a dense voxel representation network for implicit scene completion. First, we design a Bird’s-Eye View (BEV) assisted enhancement module to enhance non-empty voxel features by incorporating the information contained in the learned dense BEV features into them through deformable cross-attention. Second, we construct a feature adaptive completion module to adaptively complete voxel features using deformable self-attention, realizing the transfer of the information from non-empty voxels to empty voxels. Extensive experiments on SemanticKITTI and SemanticPOSS datasets demonstrate our method achieves state-of-the-art performance.

关键词： Point cloud compression Representation learning Fuses Convolution

来源：评论

学校读者我要写书评

暂无评论

Robust pre-departure scheduling for a nation-wide air traffic flow management

引用

Chinese Journal of Aeronautics 2025年第4期38卷 484-500页

作者： Jianzhong YAN Haoran HU Yanjun WANG Xiaozhen MA Minghua HU Daniel DELAHAYE Sameer ALAM College of Civil Aviation Nanjing University of Aeronautics and AstronauticsNanjing 210016China Operations Management Center Air Traffic Management Bureau of CAACBeijing 100018China State Key Laboratory of Air Traffic Management System Nanjing 210016China SIPSG Information Technology Co.Ltd. Suzhou 215127China ENAC Research Lab Ecole Nationale de l’Aviation CivileToulouse 31400France Air Traffic Management Research Institute Nanyang Technological UniversitySingapore 637460Singapore

Air traffic flow management has been a major means for balancing air traffic demandand airport or airspace capacity to reduce congestion and flight ***,unpredictable fac-tors,such as weather and equipment malfunctions,can cause dynamic changes in airport and sectorcapacity,resulting in significant alterations to optimized flight schedules and the calculated pre-departure ***,taking into account capacity uncertainties is essential to create a moreresilient flight *** paper addresses the flight pre-departure sequencing issue and intro-duces a capacity uncertainty model for optimizing flight schedule at the airport network *** goal of the model is to reduce the total cost of flight delays while increasing the robustnessof the optimized schedule.A chance-constrained model is developed to address the capacity uncer-tainty of airports and sectors,and the significance of airports and sectors in the airport network isconsidered when setting the violation *** performance of the model is evaluated usingreal flight data by comparing them with the results of the deterministic *** development ofthe model based on the characteristics of this special optimization mechanism can significantlyenhance its performance in addressing the pre-departure flight scheduling problem at the airportnetwork level.

关键词： Air traffic flow management Airport and airspace network Capacity uncertainty Chance constraint Stochastic optimization

来源：评论

学校读者我要写书评

暂无评论

Efficient Reinforcement Learning for 3D LiDAR Navigation of Mobile Robot 41

Efficient Reinforcement Learning for 3D LiDAR Navigation of ...

引用

第41届中国控制会议

作者： Yu Zhai Zhe Liu Yanzi Miao Hesheng Wang China University of Mining and Technology MoE Key Lab of Artificial Intelligence Shanghai Jiao Tong University Department of Automation Key Laboratory of System Control and Information Processing of Ministry of EducationShanghai Engineering Research Center of Intelligent Control and ManagementShanghai Jiao Tong University

ISBN: (数字)9789887581536

ISBN: (纸本)9781665482561

Developing an efficient automatic navigation system for mobile robots is challenging in the strange scenarios where robots can only observe the environment of the surrounding limited *** other distributed automatic navigation systems exist,they often require extracting semantic information to calculate navigation action,which requires extra modules to provide perceptual information and is not *** propose an end-to-end automatic navigation system based on the reinforcement learning *** particular,the raw 3 D LiDAR data is used to directly map an efficient navigation *** design a novel dense reward function to handle the reward sparsity issue and provide a graphical representation method to enable the efficient feature learning from the raw 3 D LiDAR data in our navigation *** addition,an imitation learning based policy initialization is introduced before the subsequent reinforcement learning,which increases the learning efficiency and,in the meantime,still encouraging the robot to explore all the potential states to achieve advanced performance than the imitated *** navigation model is trained in the Webots environment and the experimental results show that our model has efficient and flexible navigation performance in complex *** importantly,trained model can be easily extended to unfamiliar environments.

关键词： End-to-end Navigation Mobile robot Reinforcement Learning 3D LiDAR

来源：评论

学校读者我要写书评

暂无评论

Attention-Enhanced Deep Reinforcement Learning for Electric Vehicle Routing Optimization

引用

IEEE Transactions on Transportation Electrification 2025年

作者： Wang, Chao Zhang, Renyuan Hong, Rentao Wang, Haibo Anhui University State Key Laboratory of Opto-Electronic Information Acquisition and Protection Technology Engineering Research Center of Autonomous Unmanned System Technology Ministry of Education School of Artificial Intelligence Hefei230601 China University of Science and Technology of China Institute of Artificial Intelligence Hefei Comprehensive National Science Center School of Computer Science and Technology Hefei230601 China

Recently, the electric vehicle routing problem (EVRP) has emerged as a significant challenge in the transportation industry. Unlike traditional vehicle routing problems (VRPs) that focus primarily on route optimization for customers, the EVRP must also consider electricity constraints and charging decisions. Although deep reinforcement learning (DRL), particularly with attention-based Transformer architecture, has achieved success in VRPs, it faces significant difficulties when directly applied to the EVRP due to the additional multifaceted constraints and node types. To address these challenges, this paper proposes an attention-enhanced DRL method specifically designed for the EVRP. First, a heterogeneous attention mechanism is developed in the encoder to capture the intrinsic access relationships between customers and charging stations. Then, an embedding enhancement module is integrated to enrich node embeddings by incorporating adjacent distance and solution structure information through a global-local attention mechanism. Finally, a problem-size sensor is incorporated within the decoder to quantify route planning progress, thereby enhancing adaptability to the EVRP constraints. Empirical results demonstrate that the proposed method not only achieves higher efficiency compared to heuristic methods, but also significantly outperforms representative DRL-based approaches in terms of solution quality. © 2025 IEEE.

关键词： Vehicle routing

来源：评论

学校读者我要写书评

暂无评论

Study on Thermal Stability and Space Charge Characteristics of Epoxy/PMIA Fiber Composites

Study on Thermal Stability and Space Charge Characteristics ...

引用

2022 IEEE International Conference on High Voltage Engineering and Applications, ICHVE 2022

作者： Yin, Fei Yang, Lijun Hou, Wei Yuan, Yuan State Key Lab of Power Transmission Equipment & System Security and New Technology Chongqing China School of Materials Science and Engineering Chongqing University Chongqing China

ISBN: (数字)9781665407502

ISBN: (纸本)9781665407502

Aramid fiber insulation poles are widely used because of their excellent mechanical and electrical properties, and the type of curing agent plays an important role in the performance of insulation poles. In the work, the effects of different curing agents (tetrahydrophthalic anhydride (THPA), methyltetrahydrophthalic anhydride (MHAC) and methylnadic anhydride (MNA)) on the thermal stability and space charge characteristics of epoxy resin/PMIA composite structures were investigated based on molecular simulation calculations and experimental methods. Compared with other curing agent/epoxy systems, MHAC/epoxy (EPMH) systems has lower coefficient of thermal expansion, and it has stronger binding ability to PMIA. Under the action of the electric field, the energy gap between the highest electron-occupied orbital and the lowest empty orbital of the EPMH system is larger. The EPMH/PMIA composite system has less space charge accumulation under the electric field, and the charge dissipation rate of the material is faster when the electric field is removed. © 2022 IEEE.

关键词： Electric fields

来源：评论

学校读者我要写书评

暂无评论

A direct spino-cortical circuit bypassing the thalamus modulates nociception

引用

细胞研究（英文版） 2023年第10期33卷 775-789页

作者： Bing Cai Dan Wu Hong Xie Yan Chen Huadong Wang Sen Jin Yuran Song Anan Li Shiqi Huang Sashuang Wang Yingjin Lu Lan Bao Fuqiang Xu Hui Gong Changlin Li Xu Zhang Guangdong Institute of Intelligence Science and Technology HengqinZhuhaiGuangdongChina SIMR Joint Lab of Drug Innovation Shanghai Advanced Research InstituteChinese Academy of Sciences(CAS) Xuhui Central Hospital ShanghaiChina Research Unit of Pain Medicine Chinese Academy of Medical SciencesHengqinZhuhaiGuangdongChina Institute of Neuroscience and State Key Laboratory of Neuroscience Center for Excellence in Brain Science and Intelligence TechnologyCASShanghaiChina SIMR Joint Lab of Drug Innovation Shanghai Advanced Research InstituteChinese Academy of Sciences(CAS) Xuhui Central Hospital ShanghaiChina Institute of Photonic Chips School of Medical Instrument and Food Engineering University of Shanghai for Science and TechnologyShanghaiChina Shenzhen Key Laboratory of Viral Vectors for Biomedicine Brain Cognition and Brain Disease InstituteShenzhen Institute of Advanced TechnologyCASShenzhenGuangdongChina HUST-Suzhou Institute for Brainsmatics JITRI Institute for BrainsmaticsSuzhouJiangsuChina SIMR Joint Lab of Drug Innovation Shanghai Advanced Research InstituteChinese Academy of Sciences(CAS) Xuhui Central Hospital ShanghaiChina School of Life Science and Technology ShanghaiTech UniversityShanghaiChina Department of Pain Medicine and Shenzhen Municipal Key Laboratory for Pain Medicine Shenzhen Nanshan People's HospitalShenzhenGuangdongChina Guangdong Institute of Intelligence Science and Technology HengqinZhuhaiGuangdongChina Guangdong Institute of Intelligence Science and Technology HengqinZhuhaiGuangdongChina School of Life Science and Technology ShanghaiTech UniversityShanghaiChina State Key Laboratory of Cell Biology Shanghai Institute of Biochemistry and Cell BiologyCenter for Excellence in Molecular Cell ScienceCASShanghaiChina Guangdong Institute of Intelligence Science and Technology HengqinZhuhaiGuangdongChina Department of Pain Medicine and Shenzhen Municipal Key Laboratory for Pain Medicine Shenzhen Nanshan People's HospitalShenzhenGuangdongChina Guang

Nociceptive signals are usually transmitted to layer 4 neurons in somatosensory cortex via the spinothalamic-thalamocortical *** layer 5 corticospinal neurons in sensorimotor cortex are reported to receive the output of neurons in superficial layers;and their descending axons innervate the spinal cord to regulate basic sensorimotor ***,we show that a subset of layer 5 neurons receives spinal inputs through a direct spino-cortical circuit bypassing the thalamus,and thus define these neurons as spino-cortical recipient neurons(SCRNs).Morphological studies revealed that the branches from spinal ascending axons formed a kind of disciform structure with the descending axons from SCRNs in the basilar pontine nucleus(BPN).Electron microscopy and calcium imaging further confirmed that the axon terminals from spinal ascending neurons and SCRNs made functional synaptic contacts in the BPN,linking the ascending sensory pathway to the descending motor control ***,behavioral tests indicated that the spino-cortical connection in the BPN was involved in nociceptive *** vivo calcium imaging showed that SCRNs responded to peripheral noxious stimuli faster than neighboring layer 4 cortical neurons in awake *** activities of SCRNs could modulate nociceptive ***,this direct spino-cortical circuit represents a noncanonical pathway,allowing a fast sensory-motor transition of the brain in response to noxious stimuli.

关键词： Biological techniques Cell biology

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：