Computing for Embedded Lightweight Learning

EE Department @ POSTECH

We design the future learning technology and platform.

Our Research
slider-image
About Us
Primary Research Theme

Efficient AI Technology

We focus on how to redesign AI and learning technologies towards superior computing efficiency for IoT/Big Data/edge computing. We explore alternative computing solutions for future learning technology, including near data computing to push computation beyond traditional processors, and brain-inspired hyperdimensional computing that closely models the ultimate efficient processor - the human brain.
Know More

Latest News

Accepted Paper at ICML 2026 (Spotlight)

1 paper accepted: FOCUS & RePAIR, a token-level guidance framework for mitigating text degeneration in pruned large language models (ICML 2026 Spotlight).

Accepted Paper at EuroSys 2026

1 paper accepted: HVR, a hyperdimensional-computing-based framework for scalable million-scale text-to-video retrieval (EuroSys 2026).

Accepted Paper at DATE 2026, ICLR 2026

3 papers accepted: MeshHD: Near-Linear Encoding for Hyperdimensional Computing via Multi-Scale Bases and Kronecker Factorization (DATE 2026); Enhanced CXL Pooled Memory System for Scalable AI via Embedding Access Prediction (DATE 2026); GlowQ: Group-Shared Low-Rank Approximation for Quantized LLMs (ICLR 2026).

Active Research Topics

On-Device Generative AI

We advance state-of-the-art Generative AI by developing sophisticated models, including LLM and Diffusion, and applying them to emerging fields such as system optimization and performance evaluation. Our work emphasizes the development of techniques for model optimization and knowledge compression, facilitating efficient deployment across diverse application domains and driving innovation in modern system design.

Brain-inspired Hyperdimensional Computing

Hyperdimensional (HD) computing is an alternative computing method that processes cognitive tasks in a lightweight and error-torrent way based on theoretical neuroscience. We work on developing neuro-symbolic AI to support various learning tasks such as supervised, unsupervised, and reinforcement learning based on HD computing.

Learning with Alternative Computing

We rethink the role of machine learning in various alternative computing. We explore diverse system-level solutions to seamlessly accelerate AI applications on next-generation computer architectures such as near-data computing, in-memory computing, and CXL (Compute Express Link). We also develop ML-driven system software to enhance the capability of traditional computers.

Address

LG Research Building #301, 77 Cheongam-Ro, Namgu, Pohang (Postal code: 37673), REPUBLIC OF KOREA

Email

yeseongkim@postech.ac.kr

Phone

(+82) 054-279-2218