Track 3 | Special Track 专题分会场
Large Model Inference and Edge Intelligence: Cache Management and Storage Optimization
大模型推理与边缘智能：缓存管理与存储优化
As AI applications expand in IoT and smart cities, resource-constrained deployment, real-time response, data transfer, and energy consumption present major challenges. This track focuses on large model inference and edge intelligence, including KV cache management and reuse, memory and storage coordination, inference scheduling, model compression, and device-edge-cloud collaboration. We welcome research on efficient deployment of large models and other AI models on devices, edge platforms, and the cloud, covering algorithm and system optimization, hardware-software co-design, performance evaluation, and practical applications that balance latency, throughput, accuracy, cost, and energy consumption.
随着人工智能在物联网、智慧城市等场景中的应用不断深入，资源受限部署、实时响应、数据传输和能耗控制成为智能计算系统的重要挑战。本专题聚焦大模型推理与边缘智能，重点关注 KV 缓存管理与复用、内存与存储协同、推理调度、模型轻量化及端边云协同计算。欢迎大模型及其他人工智能模型在终端、边缘和云端的高效部署研究，涵盖算法与系统协同优化、软硬件协同设计、性能评测和应用实践，探索兼顾时延、吞吐量、精度、成本与能耗的智能计算技术。
Organizers 专题组织者
Track Chair / 专题主席Jinhua Cui / 崔金华
Associate Professor, School of Computer Science and Technology, Huazhong University of Science and Technology
华中科技大学计算机科学与技术学院，副教授
jhcui@hust.edu.cn Co-chair / 专题联席Shiqiang Nie / 聂世强
Associate Research Professor, School of Computer Science and Technology, Xi'an Jiaotong University
西安交通大学计算机科学与技术学院，副研究员
shiqiang.nie@xjtu.edu.cn Co-chair / 专题联席Chunfeng Du / 杜春锋
Associate Research Professor, School of Computer Science and Artificial Intelligence, Zhengzhou University
郑州大学计算机与人工智能学院，副研究员
dcf@zzu.edu.cn Track Topics 征稿主题
Topics include, but are not limited to, the following:
本专题欢迎以下方向的研究投稿，包括但不限于：
- KV cache allocation, reuse, eviction, and lifecycle management.
KV缓存的分配、复用、淘汰与生命周期管理。
- KV cache quantization, compression, sparsification, and accuracy-performance trade-offs.
KV缓存量化、压缩、稀疏化及精度与性能权衡。
- Hierarchical caching, offloading, and prefetching across GPU memory, main memory, and SSDs.
GPU显存、主存和SSD之间的分层缓存、数据卸载与预取。
- Inference systems for long contexts, retrieval-augmented generation (RAG), and AI agents.
长上下文、检索增强生成（RAG）和智能体的推理系统。
- Prefill-decode disaggregation, KV cache transfer, and distributed inference serving.
Prefill与Decode分离、KV缓存传输和分布式推理服务。
- Cache-aware request scheduling, load balancing, and multi-tenant resource management.
缓存感知的请求调度、负载均衡和多租户资源管理。
- Large model deployment and hardware-software co-optimization on device-edge-cloud and heterogeneous platforms.
端边云及异构平台的大模型部署和软硬件协同优化。
- Performance, energy consumption, and reliability evaluation of inference systems, with IoT and smart city applications.
推理系统的性能、能耗、可靠性评测，以及物联网和智慧城市应用实践。
- Model compression, inference, and heterogeneous acceleration on resource-constrained devices and edge platforms.
资源受限终端和边缘设备上的模型轻量化、推理与异构加速。
- Device-edge-cloud collaborative inference, computation offloading, task partitioning, and resource scheduling.
端边云协同推理、计算卸载、任务划分与资源调度。
- Edge intelligence for IoT and smart cities, model and data caching, and performance and energy optimization.
面向物联网和智慧城市的边缘智能应用、模型与数据缓存、性能及能耗优化。
Invited Speakers 邀请报告人
To be arranged; the speaker list will be announced later.
视情况安排，名单后续补充。