会议日程安排
第一天 12月27日
地点：吉林大学前卫南区敬信报告厅
09:00-09:10
开场仪式
09:10-09:50
特邀报告一
Qwen & Wan 大模型背后的系统创新：训练、推理与服务的全栈挑战
钱正平，阿里云智能AI系统研究与战略总监
09:50-10:30
特邀报告二
异构融合操作系统技术挑战和架构思考
郭寒军，华为异构融合OS首席架构师
10:30-11:00
新星/优博奖颁奖及上午茶歇（30分钟）
Session S1—内存系统与操作系统
11:00-12:15（75分钟）
- CortenMM: Efficient Memory Management with Strong Correctness Guarantees (#6)
- Object-Aware Memory Compression for Smartphones (#19)
- How to Copy Memory? Coordinated Asynchronous Copy as a First-Class OS Service (#37)
- AlloyStack: A Library Operating System for Serverless Workflow Applications (#77)
- SeqAss: Conflict-Resilient Last-Level Cache against Side-Channel Attacks (#113)
12:15-14:00
午餐 湖畔餐厅智慧食堂
Session S2—AI训练与大模型系统
14:00-15:30（90分钟）
- CCL-D:A High-Precision Diagnostic System for Large-Scale Model Training (#1)
- Efficient and Adaptable Overlapping for Computation and Communication (#8)
- AutoHAAP:Automated Heterogeneity-Aware Asymmetric Partitioning for LLM Training (#14)
- Neuralink: Fast LLM Inference on Smartphones (#91)
- PAT:Accelerating LLM Decoding via Prefix-Aware Attention (#116)
- FlexPipe:Maximizing Training Efficiency for Transformer Models (#300)
15:30-16:00
下午茶歇（30分钟）
Session S3—存储系统与数据库
16:00-17:10（75分钟）
- FalconFS:Distributed File System for Large-Scale Deep Learning Pipeline (#42)
- DumpKV: Learning-Based Lifetime-Aware Garbage Collection for LSM-tree (#36)
- Mitigating the Impedance Mismatch between Prediction Query Execution and Database Engines (#69)
- OceanBase Unitization:Building the Next Generation of Online Map Applications (#84)
- KVCache Cache in the Wild:Characterizing and Optimizing KVCache Cache at a Large Cloud Provider (#96)
产业论坛
17:15-18:15（60分钟）
- 大模型长上下文微调中的高效动态数据调度，沈雯婷（阿里云）
- 鸿蒙Web技术栈的演进以及挑战，王佐（华为）
- Beyond Autoregression: Diffusion LLM 推理加速的现在与未来，郑达（蚂蚁技术研究院）
- 腾讯算力平台联邦调度架构设计与实践，胡子千（腾讯）
18:30-20:00
晚宴 莘子园三楼
20:30
执委会闭门会议 王湘浩楼 A117
第二天 12月28日
地点：吉林大学前卫南区敬信报告厅
09:00-09:40
特邀报告三
一个软工人的辩白
蒋炎岩，南京大学计算机学院副教授
09:40-10:20
特邀报告四
系统方向的读博攻略
张焕晨，清华大学交叉信息研究班（姚班）助理教授
10:20-10:50
上午茶歇（30分钟）
Session S4—图系统与向量检索
10:50-12:05（75分钟）
- OdinANN:Direct Insert for Stable Billion-Scale Vector Search (#49)
- APERTURE:Algorithm-System Co-Optimization for Temporal Graph Network Inference (#74)
- Optimizing Data Acquisitions in Multi-Robot Systems (#114)
- Query-Aware Path Inference from Spatial Videos (#195)
- FlashANNS:GPU-Driven I/O Pipelining for Billion-ScaleSimilarity Search(#223)
12:05-13:30
午餐 湖畔餐厅智慧食堂
Session S5—加速器与异构计算架构
13:30-14:30（60分钟）
- EARTH:An Efficient MoE Accelerator with Entropy-Aware Speculative Prefetch (#46)
- LightDSA:Enabling Efficient DSA Through Hardware-Aware Transparent Optimization (#62)
- Chimera:Transparent and High-Performance ISAX Heterogeneous Computing via Binary Rewriting (#75)
- ASIC-based Compression Accelerators for Storage Systems (#142)
14:30-14:45
中场休息（15分钟）
Session S6—近存计算与专用加速器
14:45-15:30（45分钟）
- SnakeMan:Applying Relation-Centric Notation to Optimize Data Swizzle in Modern NPUs (#82)
- PIMLex:A High-Performance Learned Index with Processing-in-Memory (#85)
- Uni-STC:Unified Sparse Tensor Core (#88)
Session S7—系统基础、安全与评测
15:30-16:15（45分钟）
- CuFHEDB: GPU-Accelerated Fully Homomorphic Encryption Database (#80)
- ccAl:A Compatible and Confidential System for AI Computing (#89)
- TraceRTL:Agile Performance Evaluation for Microarchitecture Exploration (#97)
16:15
闭幕致辞暨会议结束