Sihyeong Park

A Survey on Inference Engines for Large Language Models: Perspectives on Optimization and Efficiency

Sihyeong Park, S. Jeon, Chaelyn Lee, Seokhun Jeon et autres

Large language models (LLMs) are widely applied in chatbots, code generators, and search engines. Workload such as chain-of-thought, complex reasoning, and agent services significantly increase inference cost by invoking the model repeatedly. Optimization methods such as parallelism, compression, and caching have been …

kr (code pays fourni par la source)

Cardiotoxicity monitoring and cancer therapy-related cardiac dysfunction in a heterogeneous cancer population: A retrospective study

Sihyeong Park, Taylor Hartshorne, Zachary Mendoza, Vinh Q. Nguyen et autres

Background: Chemotherapy-related cardiotoxicity can lead to significant heart damage, at times manifesting as cancer therapy-related cardiac dysfunction (CTRCD) or a decline in left ventricular ejection fraction (LVEF) by over 10% to below 53%. Current guidelines recommend cardiovascular risk assessments for cancer patients, …

us (code pays fourni par la source)

Exploring the Trade-Offs: Quantization Methods, Task Difficulty, and Model Size in Large Language Models From Edge to Giant

Jemin Lee, Sihyeong Park, Jinse Kwon, Jihun Oh et autres

Quantization has gained attention as a promising solution for the cost-effective deployment of large and small language models. However, most prior work has been limited to perplexity or basic knowledge tasks and lacks a comprehensive evaluation of recent models like Llama-3.3. In …

kr, ua (code pays fourni par la source)

A Survey on Inference Engines for Large Language Models: Perspectives on Optimization and Efficiency

Sihyeong Park, S. Jeon, C. H. Lee, Byung‐Soo Kim

Large language models (LLMs) are widely applied in chatbots, code generators, and search engines. Workload such as chain-of-throught, complex reasoning, agent services significantly increase the inference cost by invoke the model repeatedly. Optimization methods such as parallelism, compression, and caching have been …

Mixed Non-linear Quantization for Vision Transformers

Jemin Lee, Sihyeong Park, Yongin Kwon, Hyungshin Kim

kr (code pays fourni par la source)

Cardiotoxicity Monitoring and Cancer Therapy-Related Cardiac Dysfunction in a Heterogeneous Cancer Population: A Retrospective Study

Sihyeong Park, Taylor Hartshorne, Zachary Mendoza, Vinh Sang Nguyen et autres

us (code pays fourni par la source)

Mixed Non-linear Quantization for Vision Transformers

Gihwan Kim, Jemin Lee, Sihyeong Park, Yongin Kwon

The majority of quantization methods have been proposed to reduce the model size of Vision Transformers, yet most of them have overlooked the quantization of non-linear operations. Only a few works have addressed quantization for non-linear operations, but they applied a single …

Q-HyViT: Post-Training Quantization of Hybrid Vision Transformers With Bridge Block Reconstruction for IoT Systems

Jemin Lee, Yongin Kwon, Sihyeong Park, Misun Yu et autres

Recently, vision transformers (ViTs) have superseded convolutional neural networks in numerous applications, including classification, detection, and segmentation. However, the high computational requirements of ViTs hinder their widespread implementation. To address this issue, researchers have proposed efficient hybrid transformer architectures that combine convolutional …

kr (code pays fourni par la source)

Deferrable Task Execution Model for Reducing Memory Interference in a Real-Time Multi-Core Embedded System

Hyeoksoo Jang, Sihyeong Park, Hyungshin Kim

실시간 임베디드 시스템에 요구되는 기능이 증가함에 따라 멀티코어가 적용되고 있지만 공유 메모리로 인한 코어 간 간섭으로 태스크의 실행 시간이 증가할 수 있다. 이를 해결하기 위해 AER(Acquisition Execution Restitution) 등의 태스크 실행 모델이 제안되었다. 하지만 AER 실행 모델은 처리량을 감소시키거나 전체 실행시간이 증가하는 …

Performance Analysis of Deep Learning Accelerator for Edge Inference

Sihyeong Park, Yongin Kwon, Jemin Lee

엣지 장치에서 딥 러닝 기반 추론을 위해 추론 가속기가 탑재되고 있다. 딥 러닝 추론 가속기를 통해 연산 성능과 에너지 효율을 증가시킬 수 있다. 하지만 가속기에 최적화되지 않은 모델 구조와 설정을 사용하면 메모리 접근 등의 오버헤드로 인해 최적 성능을 낼 수 없다. 본 …

Real-time motion classification with efficient event stream data processing

Seokhun Jeon, Sihyeong Park, Kyung Mo Kim, Byung‐Soo Kim et autres

In this paper, we introduce an effective technique for real-time motion classification using event cameras to process input data streams. Our method allows for real-time operation. To enhance memory efficiency, our approach reduces buffers in the event-to-frame conversion process, while simultaneously distributing …

kr (code pays fourni par la source)

Soft-Error Injection System for Processor on FPGA Platform

Jea Hack Lee, Hyeonseong Kim, Sihyeong Park, Yun Pyo Hong et autres

This paper presents a soft-error injection system on the FPGA platform using an ARM Cortex-M3 processor. The system consists of an error injector that generates random numbers using a Fibonacci linear feedback shift register to introduce errors into an application program status …

kr (code pays fourni par la source)