System Infrastructure for Next-generation Data Lakehouse and Tensor Database System Research

Project Duration: May 2024 ~ April 2030

Total Grant Amount: KRW ~400,000,000 (approx. USD 400K)

Project Overview

• Infrastructure for Multimodal Learning & Databases: Establishes a distributed infrastructure (multi-node compute/storage) to support scalable multimodal learning pipelines and next-gen databases like data lakehouses and tensor DBs
• Optimization for Multimodal ML: Proposes novel in-process ML operators embedded directly within the lakehouse engine to reduce data movement and boost multimodal training over conventional pipelines
• Tensor DB with Unified Querying: Designs a novel DBMS that combines relational algebra and linear algebra for seamless querying over structured and tensor data—ideal for ML-centric analytics
• HW-SW Co-design for AI Runtimes: Implements PIM/CXL-based runtime integration and profiling (with industry prototypes) to co-optimize hardware accelerators with learning pipelines
• Cross-national Research Collaboration: Anchored in global partnerships (e.g., Univ. of Tokyo, East China Normal Univ., Microsoft, SK Hynix), with planned international workshops and A3 Foresight Program participation