• 学术动态

2026年第八届软件工程和计算机科学会议预告
作者: 发布日期:2026-04-14 浏览次数:

一、会议地点:杭州,Sniff思耐酒店(杭州西湖风景区西溪湿地公园店,西湖区灵溪北路21号)

二、主要内容:学术领域知名教授和参会者将围绕软件工程与计算机科学两大主题进行主旨报告和作者汇报

三、会议时间:2026年4月17-19

四、具体行程:

2026417日(周五)

10:00-17:00参会者签到及会议资料领取

2026418日(周六)

9:00-9:10 大会开幕式致辞

9:10-9:50 大会主讲报告1

李挥教授,北京大学,中国

报告主题:大语言模型推理中 KV 缓存瓶颈的优化

9:50-10:30 大会主讲报告2

薛雅娟教授,成都信息工程大学,中国

报告主题:基于薛定谔方程的自适应变换在地震反射数据中的应用

10:30-10:50 茶歇时间

10:50-11:30 大会主讲报告3

严善楷副教授,海南大学,中国

报告主题:面向大语言模型与知识图谱的知识海洋双向增强方法

11:30-12:00 大会主讲报告4

徐新教授,武汉科技大学,中国

报告主题:神经符号数据库技术及应用

12:00-13:00 午餐

13:00-15:30 并行分会1(作者报告)

15:30-15:50 茶歇时间

15:50-18:00 并行分会2(作者报告)

18:00-19:00 晚餐

2026419日(周日)

9:00-9:40 大会主讲报告5

董冕雄教授,室兰工业大学,日本

报告主题:粒球计算:一种高效、稳健且可解释的新型人工智能理论

9:40-10:40 并行分会3(作者报告)

10:40-11:00 茶歇时间

11:00-12:00 并行分会4(作者报告)

12:30-13:00 午餐

13:00-18:00 学术参观

18:00-19:00 晚餐

 

报告题目一:Managing the KV Cache Bottleneck in Large Language Model Inference

 告 人:李挥教授,北京大学,中国

报告时间:2026418日(周六)上午0910-0950

报告地址浙江工业大学

报告摘要:

As large language models (LLMs) increasingly underpin mission-critical applications

across industries, optimizing their inference efficiency has emerged as a critical priority. The management of the Key-Value (KV) cache, which stores the reusable computation intermediates during generation, has become the most prominent bottleneck for LLM inference optimization.

 

In this talk, we examine recent advancements in system-level and algorithmic advances in KV cache management, emphasizing (1) online approaches that dynamically allocate computational and memory resources during inference, and (2) offline strategies that precompute, structure, and compress the KV cache as the explicit memory for LLM. We evaluate techniques optimized for diverse operational contexts, spanning traditional chatbot serving and knowledge-enhanced question answering, and discuss corresponding architectural optimizations. Finally, we outline promising research directions to further address challenges in multi-instance inference. These advancements are crucial for enabling scalable enterprise solutions as LLMs expand into knowledge-enhanced, latency-sensitive, and high-throughput industrial applications.

 

报告人简介

李挥,北京大学教授、国际学术科学技术创新中心首席信息科学家、俄罗斯自然科学院外籍院士、联合国科技促进发展委员会指导下的世界数字技术学院专家委员会委员、IET会士、IEEE及中国计算机学会高级会员。他于1986年和1989年分别获得清华大学信息工程学院学士和硕士学位,2000年获香港中文大学信息工程学博士学位。曾任深圳市信息理论与未来网络架构重点实验室主任、国家重大科研基础设施——北京大学中国网络创新环境实验室主任。

 

他提出全球首个基于区块链技术的共治未来网络架构“MIN”,并在运营商网络上实现原型部署,该成果于2019年在第六届世界互联网大会(中国乌镇)获评“世界互联网领先科技成果”。曾受邀担任《ZTE通信》2020年3月刊(第18卷第1期/总第69期)“互联网域名与标识符:架构与系统”专题客座编辑。全球首部以“网络空间联合国”为主题的英文专著《共治主权网络:法律基础及其MIN架构原型与应用》已由斯普林格出版社出版。研究方向包括网络架构、网络空间安全、区块链、分布式存储。作为第一作者,在未来网络架构、区块链共识算法、分布式存储理论与系统领域已出版四部专著。

 

报告题目二:

Application of the Schr?dinger Equation–Based Adaptive Transformation to Seismic Reflection Data

 告 人:薛雅娟教授,成都信息工程大学,中国

报告时间:2026418日(周六)上午0950-1030

报告地址浙江工业大学

报告摘要:

The increasing demand for high-resolution characterization of complex geological targets in seismic reflection data necessitates advanced signal processing techniques. This study explores the filtering and denoising capabilities of a novel adaptive transformation based on the Schr?dinger equation. This transformation projects seismic data onto a basis composed of wave functions derived from the Schr?dinger equation, where each wave function corresponds to a distinct energy level and projection coefficient (PC). Seismic data can be reconstructed by combining these wave functions with their associated PCs. We introduce a quantum filtering approach that leverages this basis and dynamically adjusts PCs based on energy variations to extract localized features from seismic data. Building on this, we develop a local quantum denoising (LQD) method by integrating the filtering approach with a threshold factor. Synthetic examples demonstrate that the proposed quantum filtering method outperforms adaptive decomposition techniques like variational mode decomposition (VMD) in isolating distinct subsurface features, particularly in the presence of random near-surface velocity variations. Additionally, LQD achieves superior noise attenuation in structurally complex seismic data compared to traditional Bayesian wavelet soft thresholding and global quantum denoising (GQD). Field data applications further validate these findings. For a gas reservoir in China, quantum filtering more effectively separates strong reflections from weathered crust and gas reservoir components than VMD. For a carbonate reservoir in the same region, LQD better preserves dipping event information than Bayesian wavelet soft thresholding and GQD, highlighting its potential for enhancing seismic data interpretation in challenging geological settings. The research presented in this study was supported by the National Natural Science Foundation of China (Grant No. 42474145).

报告人简介

薛雅娟,现任中国四川成都信息工程大学通信工程学院教授、长江大学地球探测与信息技术专业兼职硕士生和博士生导师。薛教授是四川省有突出贡献的优秀专家,四川省学术技术带头人后备人选,四川省杰出青年学术和技术带头人资助计划获得者。她是四川省海外高层次留学人才、澳大利亚科廷大学勘探地球物理系国家公派访问学者(2018年)以及2016年国际埃尼奖提名者。主要从事信号处理与信息提取、地球探测与信息技术等相关算法和应用研究。

 

薛教授曾参与多项国家自然科学基金资助的国家级科研项目,目前担任若干国家级与省部级科研项目的主持人,特别包括国家自然科学基金青年项目、四川省优秀青年学术带头人资助项目等。近年来,她在国内外高水平期刊及重要学术会议上发表论文120余篇。另作为研究骨干参加国家自然科学基金联合基金项目、重点项目、面上项目、科技部国家重点研发计划子课题等国家级课题5项,中石化、中石油合作项目多项。她是IEEE高级会员, SEG高级会员,中国地球物理学会终身会员,中国通信学会会员。担任国际SCI期刊《Frontiers in Earth Science》编委,EI期刊《石油与天然气地质》青年编委。为中国石油西南油气田分公司岩石物理重点实验室学术委员。为《Surveys in Geophysics》、《Geophysics》、《Journal of Petroleum Science and Engineering》等40余家国际SCI期刊审稿专家。

 

报告题目三:

Bidirectional Enhancement with a Knowledge Ocean for Large Language Models and

Knowledge Graphs

 告 人:严善楷副教授,海南大学,中国

报告时间:2026418日(周六)上午1050-1130

报告地址浙江工业大学

报告摘要:

Large language models (LLMs), such as GPT4o and DeepSeek, are making new waves in the fields of natural language processing and artificial intelligence, due to their emergent ability and generalizability. However, LLMs are black-box models, which often fall short of capturing and accessing factual knowledge. In contrast, Knowledge Graphs (KGs), Wikipedia, and Huapu for example, are structured knowledge models that explicitly store rich factual knowledge. KGs can enhance LLMs by providing external knowledge for inference and interpretability. Meanwhile, KGs are difficult to construct and evolve by nature, which challenges the existing methods in KGs to generate new facts and represent unseen knowledge. Therefore, it is complementary to unify LLMs and KGs together and, simultaneously, leverage their advantages. In this talk, we present CHACE-KO (a Connected, Hybrid, Accommodating, Contained, and Evolving Knowledge-Ocean, https://ko.zhonghuapu.com/EN) that synergizes knowledge graphs with large language models, and performs bidirectional reasoning driven by both data and knowledge.

报告人简介

严善楷,博士,现任海南大学计算机科学与技术学院副教授、博士生导师。他于华南理工大学获得学士学位,在香港城市大学计算机科学系获得博士学位,并曾在美国国立卫生研究院BioNLP实验室担任博士后研究员。他的主要研究方向包括计算生物学、生物文本挖掘、生物信息学、数据挖掘、图神经网络和深度学习等,在生物医学领域的知识图谱增强自然语言处理方面有深入研究。他多次参与国际学术交流,如担任2024年第七届机器学习和自然语言处理国际会议(MLNLP 2024)主讲嘉宾,并指导“智能安全远程办公系统”创新训练项目。作为学院国际合作办学负责人,他积极参与党的二十届三中全会精神专题学习与人才培养工作。

 

严教授曾担任《计算与结构生物技术杂志》“计算生物学与生物信息学中的生成式AI”特刊客座编辑。兼任ISMSI 2024、ISCMI 2024联合主席,以及IEEE UIC 2022、IEEE ICPADS 2023分会主席。同时担任CACML、BIBM、ICCBB、ICONIP、ACL BioNLP研讨会等多个国际会议程序委员会委员,并为ACL、BIGCOM、ICONIP、BIBM、ICHI等国际会议及《BIB》《Bioinformatics》《PlosCB》《JAMIA》等数十种权威期刊承担同行评议工作。

 

报告题目四:

Neuro-Symbolic Database Technology and Applications

 告 人:徐新教授,武汉科技大学,中国

报告时间:2026418日(周六)上午1130-1200

报告地址浙江工业大学

报告摘要:

As an emerging database platform, the Neuro-Symbolic Database deeply integrates

core technologies of artificial intelligence analysis, database querying, and big data

computing. Its prominent advantage lies in not only efficiently processing heterogeneous data (including structured and semi-structured data) but also effectively managing and parsing unstructured data such as images, videos, and texts, thereby enabling in-depth mining of multi-source heterogeneous data. This presentation will delve into the application potential of neuro-symbolic databases in handling diverse complex data types and patterns, aiming to reveal their broad application prospects in data-intensive fields. Additionally, it will elaborate on the technical challenges faced by neuro-symbolic databases and our proposed solutions, including query languages,

processing algorithms, and storage schemes for such databases. Overall, centering on

the analysis of key technical issues in neuro-symbolic databases, this presentation

strives to provide robust theoretical support and practical recommendations for the

further research and development of this field.

报告人简介

徐新,香涛学者特聘教授、湖北省青年科技晨光计划人选。上海交通大学博士后、博士、学士,新加坡南洋理工大学CSC访问学者。现为武汉科技大学计算机科学与技术学院科研副院长、教授、博士生导师。徐教授主要从事计算机视觉与人工智能领域的研究,现为IEEE Senior Member。他曾入选湖北省青年科技晨光计划,获湖北省第四届高校青年教师教学竞赛奖,并多次担任国际学术会议主席或联合主席。作为博士生导师,徐新教授培养的学生在ACM国际多媒体会议等顶级会议上发表多篇论文,其领衔的视觉人工智能研究组在计算机视觉领域取得显著成果。

 

此外,徐教授还担任过SCI国际期刊编委,国际会议大会主席;担任教育部相关人才计划的评审专家、国家留学基金委相关项目的评审专家;担任多个省市的科技成果奖励、重点研发项目、重点项目指南的评审专家。长期从事人工智能和多媒体内容分析等领域的教学科研与成果转化工作,在面向夜间低照度场景的目标检测和行人重识别方向深入开展了相关研究。在CCF-A类、IEEE/ACM Trans.、中科院SCI一区等国内外期刊上发表论文160余篇,授权专利及软著三十余项;成果入选CCF-A类国际会议ICML亮点论文、CCF-B类国际会议最佳论文提名奖、CAA-A类国际期刊封面论文。先后主持国家自然科学基金4项,获湖北省科技进步二等奖2项。

 

报告题目五:

Granular-Ball Computing: An Efficient, Robust, and Interpretable Novel Artificial Intelligence Theory

 告 人:董冕雄教授,室兰工业大学,日本

报告时间:2026419日(周日)上午0900-0940

报告地址浙江工业大学

报告摘要:

Current artificial intelligence methods predominantly rely on the finest-grained pixel-level or single-granularity representations, lacking the innate efficiency, robustness, and interpretability of human cognitive mechanisms, which prioritize large-scale perception first. To address this, Granular-Ball Computing has been proposed and developed based on multi-granularity cognitive computation theory. This approach mimics the human brain’s “large-scale-first” cognitive process by adopting a coarse-to-fine generation strategy. It uses granular-balls of varying sizes to cover data samples, enabling adaptive and efficient multi-granularity representations. By constructing a novel computational paradigm centered on granular-balls, it achieves superior efficiency, robustness, and interpretability compared to traditional AI methods. Granular-Ball Computing has garnered significant attention from renowned domestic scholars and inspired follow-up research by experts at internationally leading institutions such as the University of Michigan, Indian Institutes of Technology, and University of Alberta, demonstrating substantial potential for future development. This lecture will present key research outcomes and recent advancements in Granular- Ball Computing, including: Granular-Ball Classifier, Granular-Ball Fuzzy Sets, Granular- Ball Clustering, Granular-Ball Graph Learning, Granular-Ball Reinforcement Learning, Granular-Ball Large Models, Granular-Ball Evolutionary Computation, Granular-Ball Rough Sets, Granular-Ball Computer Vision, Granular-Ball Natural Language Processing and so on.

报告人简介

董冕雄,日本国立室兰工业大学副校长、日本工程院外籍院士,日本华侨华人博士协会副会长。2000年代在日本公立会津大学获计算机理工学学士、硕士及博士学位,现任室兰工业大学信息与电子工程系教授、博士生导师。研究领域涵盖大数据、云计算、边缘计算、5G移动网络及人工智能。

 

董教授累计发表学术论文490余篇,其中7篇入选ESI热点论文,33篇为ESI高被引论文,谷歌学术引用量超15000次。研究成果获2019年度北海道科学技术奖、2021年度日本文部科学大臣表彰青年科学家奖。主持日本学术振兴会科研项目十余项,担任《IEEE Transactions on Green Communications and Networks》等10余个国际期刊编委,参与组织IJCAI、AAAI等100余个国际学术会议。2024年在中国开展学术交流活动,主题涉及智慧医疗及产学研合作领域。