박성용 연구실

System-level optimizations for LLM and RAG serving including infrastructure-centric optimization, memory management, heterogeneous resource utilization, and QoS-aware adaptive scheduling for latency-stable serving.

Software optimization for distributed deep learning including Graph Neural Networks (GNN), model and data parallelism, vision, language, and graph models in high-performance cluster computing environments.

Vector database architecture for large-scale vector datasets with storage-based approximate nearest neighbor (ANN) search, index partitioning, and integration with stream processing engines.

LSM-tree optimization addressing write amplification, read overhead, and space amplification through compaction policies, level structure redesign, and hardware acceleration with GPUs and FPGAs.

Research on Compute Express Link (CXL) technology for memory expansion and accelerators, focusing on memory hierarchy optimization for GNN-based recommendation systems in disaggregated computing environments.

Object storage design optimized for AI and machine learning workloads, supporting unstructured data management, data lakes, and cloud-native applications with metadata-driven retrieval.

In-memory streaming processing platform optimization for heterogeneous computing infrastructures including query execution plans, task scheduling, and state management for real-time data analysis.