检索结果-内蒙古大学图书馆

Accelerating BERT inference with GPU-efficient exit prediction

Frontiers of Computer Science 2024年第3期18卷 31-42页

作者： Lei LI Chengyu WANG Minghui QIU Cen CHEN Ming GAO Aoying ZHOU Shanghai Engineering Research Center of Big Data Management School of Data Science and EngineeringEast China Normal UniversityShanghai 200062China Alibaba Group Hangzhou 311121China KLATASDS-MOE School of StatisticsEast China Normal UniversityShanghai 200062China

BERT is a representative pre-trained language model that has drawn extensive attention for significant improvements in downstream Natural Language Processing(NLP)*** complex architecture and massive parameters bring BERT competitive performance but also result in slow speed at model inference *** speed up BERT inference,FastBERT realizes adaptive inference with an acceptable drop in accuracy based on knowledge distillation and the early-exit ***,many factors may limit the performance of FastBERT,such as the teacher classifier that is not knowledgeable enough,the batch size shrinkage and the redundant computation of student *** overcome these limitations,we propose a new BERT inference method with GPU-Efficient Exit Prediction(GEEP).GEEP leverages the shared exit loss to simplify the training process of FastBERT from two steps into only one step and makes the teacher classifier more knowledgeable by feeding diverse Transformer outputs to the teacher *** addition,the exit layer prediction technique is proposed to utilize a GPU hash table to handle the token-level exit layer distribution and to sort test samples by predicted exit *** this way,GEEP can avoid batch size shrinkage and redundant computation of student *** results on twelve public English and Chinese NLP datasets prove the effectiveness of the proposed *** source codes of GEEP will be released to the public upon paper acceptance.

关键词： BERT FastBERT inference acceleration model distillation early exit text classification

来源：评论

学校读者我要写书评

暂无评论

UnifiedGEC: Integrating Grammatical Error Correction Approaches for Multi-languages with a Unified Framework 31

UnifiedGEC: Integrating Grammatical Error Correction Approac...

引用

31st International Conference on Computational Linguistics, COLING 2025

作者： Zhao, Yike Wang, Xiaoman Lan, Yunshi Qian, Weining School of Data Science and Engineering East China Normal University Shanghai Engineering Research Center of Big Data Management China

ISBN: (纸本)9798891761988

Grammatical Error Correction is an important research direction in NLP field. Although many models of different architectures and datasets across different languages have been developed to support the research, there is a lack of a comprehensive evaluation on these models, and different architectures make it hard for developers to implement these models on their own. To address this limitation, we present UnifiedGEC, the first open-source GEC-oriented toolkit, which consists of several core components and reusable modules. In UnifiedGEC, we integrate 5 widely-used GEC models and compare their performance on 7 datasets in different languages. Additionally, GEC-related modules such as data augmentation, prompt engineering are also deployed in it. Developers are allowed to implement new models, run and evaluate on existing benchmarks through our framework in a simple way. Code, documents and detailed results of UnifiedGEC are available at https://***/AnKate/UnifiedGEC. ©2025 Association for Computational Linguistics.

关键词： Error correction

来源：评论

学校读者我要写书评

暂无评论

Large Language Models are Good Annotators for Type-aware data Augmentation in Grammatical Error Correction 31

Large Language Models are Good Annotators for Type-aware Dat...

引用

31st International Conference on Computational Linguistics, COLING 2025

作者： Li, Xinyuan Lan, Yunshi East China Normal University China Shanghai Engineering Research Center of Big Data Management China

ISBN: (纸本)9798891761964

Large Language Models (LLMs) have achieved outstanding performance across various NLP tasks. Grammatical Error Correction (GEC) is a task aiming at automatically correcting grammatical errors in text, but it encounters a severe shortage of annotated data. researchers have tried to make full use of the generalization capabilities of LLMs and prompt them to correct erroneous sentences, which however results in unexpected over-correction issues. In this paper, we rethink the role of LLMs in GEC tasks and propose a method, namely TypeDA, considering LLMs as the annotators for type-aware data augmentation in GEC tasks. Different from the existing data augmentation methods, our method prevents in-distribution corruption and is able to generate sentences with multi-granularity error types. Our experiments verify that our method can generally improve the GEC performance of different backbone models with only a small amount of augmented data. Further analyses verify the high consistency and diversity of the pseudo data generated via our method. Our code can be accessed via the provided URL. © 2025 Association for Computational Linguistics.

关键词： Computational linguistics

来源：评论

学校读者我要写书评

暂无评论

Combining permissioned blockchain and Bayesian best-worst method for transparent supplier selection in supply chain management

引用

Science China(Technological Sciences) 2024年第8期67卷 2579-2593页

作者： LIU JiaJun ZHANG Jie LENG JieWu Shanghai Engineering Research Center of Industrial Big Data and Intelligent System Institute of Artificial IntelligenceDonghua UniversityShanghai 201620China State Key Laboratory of Precision Electronic Manufacturing Technology and Equipment Guangdong University of TechnologyGuangzhou 510006China

Supplier selection is an important business activity in order to realize the purchasing function in supply chain *** supplier selection process includes four stages,i.e.,bidding inviting,bidding,group decision-making,and results disclosure,involving the participation of manufacturing service demanders(MSDs),manufacturing service suppliers(MSSs),and ***,all the participants have raised concerns about the increased transparency in supplier ***,this study proposes a transparent supplier selection method by considering the engagement of *** this method,the Bayesian best-worst method(Bayesian BWM)is used to aggregate decision-makers'preferences into the overall optimal weights of the alternative MSSs,and the MSS with the largest weight is considered the suitable MSS for ***,blockchain is introduced to record the decision-making process information about supplier selection through a customized smart contract,where MSSs act as supervisors to supervise the decision-making process through the distributed consensus mechanism rather than directly participate in the decision-making ***,a case study of supplier selection in purchasing vibration acceleration sensors is *** result shows that the proposed method can support MSDs in selecting suitable MSS from alternative MSSs by aggregating decision-makers'preferences,and blockchain can provide credible information about the supplier selection process for MSSs,MSDs,and *** this way,the transparency of supplier selection is enhanced.

关键词： blockchain supplier selection supply chain transparency Bayesian BWM group decision-making

来源：评论

学校读者我要写书评

暂无评论

Hybrid Evaluation for Occlusion-based Explanations on CNN Inference Queries 40

Hybrid Evaluation for Occlusion-based Explanations on CNN In...

引用

40th IEEE International Conference on data engineering, ICDE 2024

作者： Ding, Guangyao Xu, Chen Qian, Weining School of Data Science and Engineering East China Normal University Shanghai200062 China Shanghai Engineering Research Center of Big Data Management China

ISBN: (纸本)9798350317152

Deep CNNs are increasingly prevalent in various application domains such as image processing. To explain a CNN prediction, it is popular to employ occlusion-based explanations (OBE). OBE helps users understand which parts of an image are important to a CNN prediction. Existing systems have explored incremental evaluation to accelerate CNN inference in OBE. However, they are oblivious that incremental evaluation does not always outperform full evaluation for certain layers. To address this issue, we propose a hybrid evaluation to efficiently interleave full and incremental evaluations during the CNN inference. Ad-ditionally, it employs a cost model to compare the overhead costs of two types of evaluations and a heuristic method to determine the efficient plan combination for common CNNs. More impor-tantly, hybrid evaluation adopts a dynamic programming-based method for attention-based CNNs. In particular, the dynamic programming-based method significantly reduces the overhead of searching for the efficient plan combination on the complex DAG structure. To demonstrate the efficiency of our techniques, we implement HyInJ, a hybrid CNN inf erence system based on PyTorch. Our experiments show that HyInf reduces execution time by up to 22% on GPU and 55% on CPU in comparison to the state-of-the-art incremental evaluation. © 2024 IEEE.

关键词： Dynamic programming

来源：评论

学校读者我要写书评

暂无评论

Psychological Counseling with Integration of Knowledge Graph and Multi-Agent Collaboration

引用

Journal of shanghai Jiaotong University (Science) 2025年 1-13页

作者： Liu, Xi Zheng, Xiaohu Cui, Weiwei College of Information Science and Technology Donghua University Shanghai201620 China Institute of artificial Intelligence Donghua University Shanghai201620 China Shanghai Engineering Center of Industrial Big Data and Intelligent System Shanghai201620 China Management College Wuzhou University Guangxi Wuzhou543003 China

Applying large language models (LLMs) to psychological counseling can reduce costs and lower barriers to access. However, issues inherent to LLMs, such as output hallucinations, knowledge lags, and response biases, limit the quality of counseling services. To address these challenges, we propose a psychological counseling method that integrates a knowledge graph (KG) and multi-agent collaboration (MAC). First, we design an ontology model and use an improved convolution residual network (CRN) to extract relational triples for constructing a psychological medicine KG. Second, we perform low-rank adaptation (LoRA) on Qwen1.5-7B-chat using the Text2Cypher training set. Finally, we design MAC prompts based on the characteristics of the retrieval-based question-answering (QA) process to facilitate the collaborative implementation of psychological medicine knowledge QA and updates. The experiments showed that the improved CRN achieved an extraction accuracy of 80%. The mean squared error of the general-domain capability of the LLM before and after LoRA was only 0.738, with a Text2Cypher score of 99.8 after LoRA. The results generated after prompting received a 70.44% approval rate compared to other methods. Thus, the proposed method can effectively enhance the intelligence, accuracy, and controllability of psychological counseling. © shanghai Jiao Tong University 2024.

关键词： Knowledge graph

来源：评论

学校读者我要写书评

暂无评论

Puzzle mode graph learning with pattern composition relationships reasoning for defect detection of printed products

引用

Journal of Manufacturing systems 2025年 81卷 34-48页

作者： Zhu, Zixun Zhang, Jie Wang, Junliang Zhang, Peng Li, Jiacheng College of Mechanical Engineering Donghua University Shanghai China Engineering Research Center of Artificial Intelligence for Textile Industry Ministry of Education Institute of Artificial Intelligence Donghua University Shanghai China Shanghai Engineering Research Center of Industrial Big Data Intelligent System Institute of Artificial Intelligence Donghua University Shanghai China College of Materials Science and Engineering Donghua University Shanghai China

Patterns are designs composed of specific elements and are widely present in various printed products, representing particular design intentions. However, due to printing errors, pattern defects are extremely common in these products, significantly impacting their visual quality and market price, especially in high-value customized products like luxury apparel, premium wallpapers and decorative tiles. Traditional detection methods struggle to provide effective judgments with conventional visual cues purely and frequently fall short due to the intricate nature of the pattern composition. To overcome this challenge, we propose a puzzle mode graph learning method capable of reasoning about pattern composition relationships. This novel detection framework simulates the logical reasoning ability of humans in assembling unordered puzzle pieces into a complete pattern, thus surpassing spatial structure limitations and enabling structural defect detection in patterns. Specifically, a parametric representation function is integrated into convolutional layers to enhance the segmentation accuracy of shape masks. Then, cross-graph semantic matching rules are developed to dynamically re-encode the adjacency matrix, enabling the construction of an attribute relationship graph that explicitly describes pattern attributes, including pattern elements, color sequences and shape positions. Moreover, the defective reasoning mechanism calculates puzzle-mode scores to decouple semantic relationships of defect features, inferring anomalous node and edge weights affecting the graph structure, thereby facilitating more precise judgments of pattern defects. Comparative experiments conducted on a real printed defect dataset validate this method. Results demonstrate its effectiveness and robustness in identifying complex pattern defects, providing essential support for appearance quality control in high-end industrial products. © 2025 The Society of Manufacturing Engineers

关键词： Computer control

来源：评论

学校读者我要写书评

暂无评论

MGCP: A Multi-View Diffusing Graphs Based Traffic Congestion Prediction for Roads Around Factory 9th

MGCP: A Multi-View Diffusing Graphs Based Traffic Congestion...

引用

19th International Conference on Advanced data Mining and Applications, ADMA 2023

作者： Zhao, Wei Shen, Yingzi Mao, Jiali Cheng, Lei East China Normal University Shanghai200062 China Shanghai Engineering Research Center of Big Data Management Shanghai China

ISBN: (纸本)9783031466762

Traffic congestion easily occurs on the roads around factories due to limited road space, which will be worsen especially when many trucks stay at the roadside of the same road. Actually the truck’s movement or staying at the road depends on the phase of its implementing cargo-loading task, e.g., the truck may strand in the road outside the factory due to waiting for loading cargoes, or move toward the road near the gate of factory when receiving loading notification. Thus, to predict traffic jams of the roads around the factory, it is necessary to consider the influence of the truck’s cargo-loading task phase on road traffic situation. However, the influences of different task phases that the trucks are on traffic situation of the roads are not the same, and such influences may change with the transition of the trucks’ task phases, which brings severe challenges for precise prediction. In this paper, we put forward a multi-view task diffusing graphs based traffic congestion prediction method for Roads around Factory, called MGCP. To capture discrepant impacts of task phases on traffic situations of the roads, we present a task phase based multi-view diffusing graphs generating method. In addition, we leverage a Markov process to denote the transition of each task phase, and build a transition matrix of task phases to extract the transited task phases in the future. Experimental results on real steel logistics data sets demonstrate that our proposed method outperforms the existing prediction approaches in terms of prediction accuracy. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.

关键词： Trucks

来源：评论

学校读者我要写书评

暂无评论

Accelerating Recommendation Inference via GPU Streams 28th

Accelerating Recommendation Inference via GPU Streams

引用

28th International Conference on database systems for Advanced Applications, DASFAA 2023

作者： Niu, Yuean Xu, Zhizhen Xu, Chen Wang, Jiaqiang East China Normal University Shanghai China Shanghai Engineering Research Center of Big Data Management Shanghai China Tencent Shenzhen China

ISBN: (纸本)9783031306365

Deep Learning based recommendation is common in various recommendation services and widely used in the industry. To predict user preferences accurately, state-of-the-art recommendation models contain an increasing number of features and various methods of feature interaction, which both lengthen inference time. We observe that the embedding lookup and feature interaction of different features in a recommendation model is independent of each other. However, current deep learning frameworks (e.g., TensorFlow, PyTorch) are oblivious to this independence, and schedule the operators to execute sequentially in a single computational stream. In this work, we exploit multiple CUDA streams to parallelize the execution of embedding lookup and feature interaction. To further overlap the processing of different sparse features and minimize synchronization overhead, we propose a topology-aware operator assignment algorithm to schedule operators to computational streams. We implement a prototype, namely StreamRec, based on TensorFlow XLA. Our experiments show that StreamRec is able to reduce latency by up to 27.8% and increase throughput by up to 52% in comparison to the original TensorFlow XLA. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Graphics processing unit

来源：评论

学校读者我要写书评

暂无评论

SAMI: A Shape-Aware Cycling Map Inference Framework for Designated Driving Service 39

SAMI: A Shape-Aware Cycling Map Inference Framework for Desi...

引用

39th IEEE International Conference on data engineering, ICDE 2023

作者： Shen, Wenyi Wu, Wenyu Mao, Jiali Chen, Jie Cao, Shaosheng Zhao, Lisheng Zhou, Aoying Zhou, Lin East China Normal University Shanghai China DiDi Chuxing China Shanghai Engineering Research Center of Big Data Management China

ISBN: (纸本)9798350322279

Along with the increase in strict regulation of drunk driving behavior in China, the demands for designated driving services have risen in popularity. In the absence of specialized cycling map for the designated drivers who use foldable electric bicycles, they tend to take a detour or are lost on the way to the car owners' appointed parking places. With gradual popularization of chauffeur services, cycling trajectories generated by designated drivers almost spread all over the city. It provides a chance for inferring the cycling map dedicated to the designated drivers. However, to infer an accurate map using trajectories faces severe challenges stemming from random cycling behaviors of designated drivers, including (i) trajectories contain a lot of noises and incomplete segments, (ii) turning trajectories at minor intersections are very sparse and (iii) trajectories on the roads of distinct shapes are obviously different. To address the above challenges, we propose a three-phase map inference framework, called SAMI, consisting of trajectory refinement, intersection pinpointing, and road curve interlinking. Specifically, cycling behavioral differences from neighbor regions are incorporated into intersection identification process to ensure obtaining high detection precision even when trajectory data is sparse. Further, shape-aware based centerline fitting strategy is put forward to guarantee that inferred road curves are consistent with real road shape as possible. Finally, extensive comparative experiments on two real data sets demonstrate that SAMI significantly outperforms state-of-the-art methods by 13.31% in F1-score of map inference and by 44.88% in recall rate of minor intersection detection. © 2023 IEEE.

关键词： Trajectories

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：