检索结果-内蒙古大学图书馆

Blockwise compression of transformer-based models without retraining

NEURAL NETWORKS 2024年 171卷 423-428页

作者： Dong, Gaochen Chen, W. TensorChip Beijing Peoples R China Tsinghua Berkeley Shenzhen Inst RISC V Int Open Source Lab Shenzhen Peoples R China

Transformer-based models, exemplified by GPT-3, ChatGPT, and GPT-4, have recently garnered considerable attention in both academia and industry due to their promising performance in general language tasks. Nevertheless, these models typically involve computationally encoding processes, and in some cases, decoding processes as well, both of which are fundamentally large-scale matrix multiplication. These operations bring the inevitable challenges of massive computation resources and huge memory footprint, usually requiring at least 1023 FLOPs and hundreds of gigabytes, respectively. A common method to address this issue is to reduce the computational and memory requirements by applying layerwise quantization to the transformer, replacing the usual fp32 data type with a low-bit equivalent. Unfortunately, this method often leads to decreased model accuracy and necessitates time-consuming retraining. Such retraining not only requires fine-tuning skills but also substantial computational resources, posing challenges for users. To specifically tackle these issues, we propose BCT, a framework of blockwise compression for transformers without retraining, aiming to facilitate model deployment. Unlike layerwise compression methods, BCT achieves finer compression of the entire transformer by operating blockwise. This method mitigates data distribution deviation caused by quantization, eliminating the requirement for retraining. BCT effectively compresses all components of the model, including but not limited to the embedding, matrix multiplication, GELU, Softmax, layer normalization, and intermediate results. In a case study, an efficient model is compressed by BCT achieving up to 7.988x compression. Subsequently, we also evaluate it on several General Language Understanding Evaluation (GLUE) datasets. Experimental results on the majority of GLUE benchmark demonstrate the effectiveness of our method, as BCT achieves less than a 0.9% degradation in accuracy compared to the more

关键词： Transformer Compression Blockwise Noretraining

来源：评论

学校读者我要写书评

暂无评论

SimLiquid: A Simulation-Based Liquid Perception Pipeline for Robot Liquid Manipulation

引用

JOURNAL OF FIELD ROBOTICS 2025年

作者： Huang, Yan Zhang, Jiawei Yu, Ran Li, Shoujie Ding, Wenbo Tsinghua Univ Shenzhen Int Grad Sch Shenzhen Ubiquitous Data Enabling Key Lab Shenzhen Peoples R China RISC V Int Open Source Lab Shenzhen Peoples R China

Transparent liquid volume estimation is crucial for robot manipulation tasks, such as pouring. However, estimating the volume of transparent liquids is a challenging problem. Most existing methods primarily focus on data collection in the real world, and the sensors are fixed to the robot body for liquid volume estimation. These approaches limit both the timeliness of the research process and the flexibility of perception. In this paper, we present SimLiquid20k, a high-fidelity synthetic data set for liquid volume estimation, and propose a YOLO-based multi-task network trained on fully synthetic data for estimating the volume of transparent liquids. Extensive experiments demonstrate that our method can effectively transfer from simulation to the real world. In scenarios involving changes in background, viewpoint, and container variations, our approach achieves an average error of 5% in real-world volume estimation. In addition, our work conducts two application experiments integrating with GPT-4, showcasing the potential of our method in service robotics. The accompanying videos and supporting Information are available at .

关键词： liquid estimation LLM Sim2Real transparent object

来源：评论

学校读者我要写书评

暂无评论

Beyond Aggregation: Efficient Federated Model Consolidation with Heterogeneity-AdaptiveWeights Diffusion 24

Beyond Aggregation: Efficient Federated Model Consolidation ...

引用

33rd ACM international Conference on Information and Knowledge Management (CIKM)

作者： Li, Jiaqi Qu, Xiaoyang Ding, Wenbo Zhao, Zihao Wang, Jianzong Ping An Technol Shenzhen Co Ltd Shenzhen Peoples R China Tsinghua Univ Shenzhen Int Grad Sch Shenzhen Peoples R China Tsinghua Univ RISC V Int Open Source Lab Shenzhen Int Grad Sch Tsinghua Berkeley Shenzhen Inst Shenzhen Peoples R China RISC V Int Open Source Lab Shenzhen Peoples R China

ISBN: (纸本)9798400704369

As the internet of Things (IoT) evolves, the need for enhanced data-sharing to improve edge device performance has led to the adoption of Federated Learning (FL) for data privacy and optimized data utilization. However, communication costs in FL remain a significant challenge. Traditional methods focus on client enhancements but overlook server-side aggregation, potentially increasing client computation loads. In response, we introduce a novel method, FedDiff, which utilizes diffusion models for generating model weights on FL servers, replacing traditional aggregation methods. Our approach, tailored for heterogeneous environments, significantly improves communication efficiency, achieving faster convergence and robust performance against weight noise in rigorous tests.

关键词： Federated Learning integration and Aggregation Diffusion Models

来源：评论

学校读者我要写书评

暂无评论

Point-Wise Vibration Pattern Production via a Sparse Actuator Array for Surface Tactile Feedback

Point-Wise Vibration Pattern Production via a Sparse Actuato...

引用

IEEE international Conference on Robotics and Automation (ICRA)

作者： Li, Xiaosa Zhao, Runze Lu, Chengyue Xiao, Xiao Ding, Wenbo Tsinghua Univ Tsinghua Shenzhen Int Grad Sch Tsinghua Berkeley Shenzhen Inst Shenzhen Peoples R China RISC V Int Open Source Lab Shenzhen 518055 Peoples R China

ISBN: (纸本)9798350384581;9798350384574

Surface vibration tactile feedback is capable of conveying various semantic information to humans via handheld electronic devices, such as smartphones, touch panels, and game controllers. However, covering the entire contacting surface of the device with a dense arrangement of actuators can affect its normal use. Determining how to produce desired vibration patterns at any contact point with only a few sparse actuators deployed on the surface of the handheld device remains a significant challenge. In this work, we develop a tactile feedback board in the size of a smartphone with only five actuators, and achieve the precise production of vibration patterns that can focus at any desired position on the board. Specifically, we investigate the vibration characteristics of a single passive coil actuator and construct its vibration pattern model for any position on the feedback board surface. Optimal phase and amplitude modulation, determined using the simulated annealing algorithm, is employed with five actuators in a sparse array. The vibration patterns from all actuators are superimposed linearly to synthetically generate different onboard vibration energy distributions for tactile sensing. Experiments demonstrated that point-wise vibration pattern production on our tactile board achieved an average level of about 0.9 in the Structural Similarity Index Measure (SSIM) evaluation, when compared to the ideal single-point-focused target vibration pattern. Four point-wise patterns focused on the top, bottom, left, and right parts of the tactile board were applied, to guide continuous directional movements without visual assistance, which shows significant implications for machine-assisted cognition based on vibration tactile feedback.

关键词： Simulated annealing

来源：评论

学校读者我要写书评

暂无评论

A hybrid wind and raindrop energy harvesting operating on Savonius turbine

引用

NANO ENERGY 2024年 127卷

作者： Sou, Kit Wa Lin, Zenan Jin, Yuchao Ping, Siqi Ding, Wenbo Wang, Jiyu Tsinghua Univ Tsinghua Berkeley Shenzhen Inst Shenzhen Int Grad Sch Shenzhen 518055 Peoples R China RISC V Int Open Source Lab Shenzhen 518055 Peoples R China Huazhong Univ Sci & Technol State Key Lab Adv Electromagnet Engn & Technol Wuhan 430074 Peoples R China

In offshore environments, abundant wind and rainfall resources await development, representing significant untapped energy potential for powering internet of Things (IoT) devices. Triboelectric Nanogenerators (TENGs) are particularly suited for capturing continuous motion in offshore environments, presenting a reliable solution for sustainable energy generation in IoT applications. Herein, we propose a wind-droplet hybrid generator (WDHG) tailored for adaptive environmental energy harvesting, explicitly targeting the demanding conditions prevalent in offshore environments. The WDHG combines the efficiency of a helical Savonius TENG with the versatility of an interdigitated electrode TENG, enabling simultaneous wind and rain energy harvesting. The helical Savonius TENG exhibits superior aerodynamic performance, producing an output power ranging from 0.39 to 2.10 mW with wind speeds increasing from 0.75 to 6.84 m/s. Additionally, the droplet TENG generates a power output of 35.90 mW/m2 under rainfall at 70 mm/min. Optimizations include not only the comparative analysis of helical Savonius turbines and the variation in pole-pair numbers but also adjustments in fluorinated ethylene propylene (FEP) film thickness and electrode gap length for interdigitated electrode, alongside the development of a tailored power management circuit (PMC) featuring dual bridge rectifiers, a buck circuit module, and a smart load switch. The PMC enables continuous operation of a low-powered digital thermohygrometer under conditions of 4.5 m/s wind speed, and drives a 20-second wake BLE server under 7.2 m/s wind speeds. integration of the WDHG with offshore wind and raindrop extends energy harvesting capabilities, exhibiting a great potential of providing reliable power source for IoT devices.

关键词： Wind energy harvesting Droplet energy harvesting Power management circuit Triboelectric nanogenerator

来源：评论

学校读者我要写书评

暂无评论

TGF-Net: Sim2Real Transparent Object 6D Pose Estimation Based on Geometric Fusion

引用

IEEE ROBOTICS AND AUTOMATION LETTERS 2023年第6期8卷 3867-3874页

作者： Yu, Haixin Li, Shoujie Liu, Houde Xia, Chongkun Ding, Wenbo Liang, Bin Tsinghua Shenzhen Int Grad Sch Shenzhen 518055 Peoples R China RISC V Int Open Source Lab Shenzhen 518055 Peoples R China Tsinghua Univ Nav & Control Res Ctr Dept Automat Beijing 100084 Peoples R China

Transparent objects are a common part of daily life, but their unique optical properties make estimating their 6D pose a challenging task. In this letter, we propose TGF-Net, a monocular instance-level 6D pose estimation method for transparent objects based on geometric fusion. TGF-Net learns the edge features and surface fragments of transparent objects as intermediate features and reduces the influence of appearance changes on the 6D pose estimation of transparent objects by fusing rich geometric features in the network. Moreover, we propose an approach for generating high-fidelity large-scale synthetic datasets of transparent objects using Blender and use this approach to generate a synthetic dataset Trans6D-32 K. Trans6D-32 K contains rendered RGB images and poses information about transparent objects in a variety of different backgrounds, perspectives, and lighting conditions. To evaluate the performance of TGF-Net on 6D pose estimation of transparent objects, we compare with multiple related works on the dataset Trans6D-32 K. TGF-Net can be trained entirely on synthetic datasets without fine-tuning and applied directly to real-world scenarios. Multiple challenging real-scene experiments demonstrate the good performance of TGF-Net, while grasping experiments demonstrate the application value of TGF-Net in transparent object manipulation.

关键词： Pose estimation Feature extraction Lighting Synthetic data Grasping Task analysis Three-dimensional displays Deep learning for visual perception data sets for robotic vision perception for grasping and manipulation

来源：评论

学校读者我要写书评

暂无评论

M³ARL: MOMENT-EMBEDDED MEAN-FIELD MULTI-AGENT REINFORCEMENT LEARNING FOR CONTINUOUS ACTION SPACE 49

M<SUP>3</SUP>ARL: MOMENT-EMBEDDED MEAN-FIELD MULTI-AGENT REI...

引用

49th IEEE international Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Tang, Huaze Hu, Yuanquan Zhao, Fanfan Yan, Junji Dong, Ting Ding, Wenbo Tsinghua Univ Shenzhen Int Grad Sch Tsinghua Berkeley Shenzhen Inst Inst Data & Informat Shenzhen 518055 Peoples R China RISC V Int Open Source Lab Shenzhen 518055 Peoples R China

ISBN: (纸本)9798350344868;9798350344851

Mean-field theory offers a promising solution to the scalability issues encountered in multi-agent reinforcement learning (MARL) within large-scale systems. However, most existing MARL algorithms based on mean-field theory are typically constrained to discrete action space. In continuous action space, the conventional mean-field approximation of meanfield theory breaks down since the representation of mean-field action is not well-defined. In this paper, we propose Moment-Embedded Mean-Field Multi-Agent Reinforcement Learning (M3ARL) for continuous action space, embedding mean-field action into multi-order moments. Specifically, we analyze the mean-field approximation on Wasserstein space and derive the approximated form of Q-function. Furthermore, to derive a learn-able form of Q-function, we apply the cylindrical function and propose the moment-embedded representations. Based on the representations and proximal policy optimization (PPO) algorithm, we propose a novel algorithm, M(3)APPO. Finally, we validate the efficacy of the proposed algorithm through experiments of predator-prey environment on continuous action space scenarios.

关键词： Deep Learning Techniques Multi-Agent Reinforcement Learning Mean Field Theory

来源：评论

学校读者我要写书评

暂无评论

FedHAP: Federated Hashing With Global Prototypes for Cross-Silo Retrieval

引用

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS 2024年第4期35卷 592-603页

作者： Yang, Meilin Xu, Jian Ding, Wenbo Liu, Yang Tsinghua Univ Tsinghua Berkeley Shenzhen Inst Tsinghua Shenzhen Int Grad Sch Beijing 100190 Peoples R China V Int Open Source Lab RISC Tsinghua Shenzhen Int Grad Sch Shenzhen 518055 Peoples R China Shanghai Artificial Intelligence Lab Shanghai 201210 Peoples R China Tsinghua Univ Inst AI Ind Res AIR Beijing 100190 Peoples R China

Deep hashing has been widely applied in large-scale data retrieval due to its superior retrieval efficiency and low storage cost. However, data are often scattered in data silos with privacy concerns, so performing centralized data storage and retrieval is not always possible. Leveraging the approach of federated learning (FL) to perform deep hashing is a recent research trend. However, existing frameworks mostly rely on the aggregation of the local deep hashing models, which are trained by performing similarity learning with local skewed data only. Therefore, they cannot work well for non-IID clients in a real federated environment. To overcome these challenges, we propose a novel federated hashing framework that enables participating clients to jointly train a shared deep hashing model by leveraging the class-wise prototypical hash codes. Globally, sharing global prototypes with only one prototypical hash code per class assists in learning consistent code distributions across clients while minimizing the cost of communication and privacy. Locally, the use of the global prototypical hash codes are maximized by jointly training a discriminator network and the local hashing network. Extensive experiments on benchmark datasets are conducted to demonstrate that our method can significantly improve the performance of the deep hashing model in the federated environments with non-IID data distributions.

关键词： Federated learning information retrieval deep hashing

来源：评论

学校读者我要写书评

暂无评论

Physical Layer Security Based on 2-D Spatial Position Optimization of Relays in Reflected Indoor VLC System

引用

IEEE WIRELESS COMMUNICATIONS LETTERS 2024年第8期13卷 2235-2239页

作者： Wu, Kun Dong, Yuhan Yu, Xinchun Song, Jian Zhang, Xiao-Ping Tsinghua Univ Shenzhen Key Lab Ubiquitous Data Enabling RISC V Internat Open Source Lab Shenzhen 518055 Peoples R China Tsinghua Univ Shenzhen Int Grad Sch Shenzhen 518055 Peoples R China Tsinghua Univ Dept Elect Engn Beijing 100084 Peoples R China

A novel relay-aided physical layer security scheme is proposed in this letter. The channel model incorporates partial reflection components in the received signal according to the signal power. With this channel model, we formulate three security capacity maximization problems based on three relay-aided scenarios: cooperative jamming, decode-and-forward, and amplify-and-forward. By using an iterative algorithm, we jointly optimize the beamforming vector and two-dimensional spatial positions of relays based on the formulated problems to maximize the sum data rate while defending against eavesdropping. Numerical results show that the sum data rate for legal receivers can be improved by applying the proposed scheme.

关键词： Receivers Relays Security interference Law Signal to noise ratio Vectors Visible light communication trusted cooperative relay physical layer security superposition coding

来源：评论

学校读者我要写书评

暂无评论

Physics-informed Neural ODE for Post-disaster Mobility Recovery 24

Physics-informed Neural ODE for Post-disaster Mobility Recov...

引用

30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

作者： Li, Jiahao Wang, Huandong Chen, Xinlei Tsinghua Univ Shenzhen Int Grad Sch Shenzhen Peoples R China Tsinghua Univ Dept Elect Engn Beijing Peoples R China Tsinghua Univ Pengcheng Lab Shenzhen Int Grad Sch RISC V Int Open Source Lab Shenzhen Peoples R China

ISBN: (纸本)9798400704901

Urban mobility undergoes a profound decline in the aftermath of a disaster, subsequently exhibiting a complex recovery trajectory. Effectively capturing and predicting this dynamic recovery process holds paramount importance for devising more efficient post-disaster recovery strategies, such as resource allocation to areas with protracted recovery periods. Existing models for post-disaster mobility recovery predominantly employ basic mathematical methods, which are strongly based on simplifying assumptions, and their limited parameters restrict their capacity to fully capture the mobility recovery patterns. In response to this gap, we introduce the Coupled Dynamic Graph ODE Network (CDGON) to model the intricate dynamics of post-disaster mobility recovery. Our model seamlessly integrates existing physical knowledge pertaining to post-disaster mobility recovery and incorporates the nuanced interactions between intra-regional and inter-regional population flows. Extensive experimental results demonstrate the efficiency of our model in capturing the dynamic recovery patterns of urban population mobility in post-disaster scenarios, surpassing the capabilities of current dynamic graph prediction models.

关键词： Physics-informed Graph Neural ODE Urban Mobility Recovery Post-Disaster

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：