Gi-Ho Park

Sejong University, Korea

Rethinking AI Acceleration: From Data to Metadata-Centric Computing

Abstract

Modern AI workloads expose fundamental limitations of conventional data-centric acceleration, driven by irregular sparsity, outliers, and memory inefficiency. 

While existing approaches focus on optimizing data movement and computation, a growing portion of system complexity is governed by metadata, including indices and value representations.
This talk presents a metadata-centric computing perspective, where metadata is treated as a control plane for AI execution. 

We illustrate this concept through case studies such as Non-Zero Bitmap (NZB) indexing and outlier-aware quantization, showing how structured metadata enables more predictable execution and improved efficiency. We conclude by discussing how this shift opens a new design space for AI accelerators and memory systems, and outline opportunities for industry–academia co-design and collaboration.

Biography

Gi-Ho Park received the B.S., M.S., and Ph.D. degrees in Computer Science from Yonsei University, Seoul, Korea, in 1993, 1995, and 2000, respectively. 

He is currently a Professor in the Department of Computer Science and Engineering at Sejong University, Korea. 

Before joining Sejong University, he worked as a Senior Engineer at Samsung Electronics in the Processor Architecture Lab, System LSI Division, from 2002 to 2008.

His research interests include computer architecture, AI accelerator design, memory and processing-in-memory (PIM) systems, system-on-chip (SoC) design, and low-power edge computing.