CV
My academic CV, covering research on efficient deep learning and large vision-language models, along with publications, awards, and service.
Contact Information
| Name | Dayoung Kil |
| Professional Title | Ph.D. Student |
| dayoungkil@soongsil.ac.kr |
Professional Summary
Ph.D. Student at the Visual Intelligence and Platform (VIP) Lab, Soongsil University, working under Prof. Seong-heum Kim. My research centers on hardware-efficient artificial intelligence — developing deep learning models that remain performant under real-world compute and latency constraints, with current emphasis on token pruning and efficient inference for large vision-language models (LVLMs) and multimodal LLMs.
Education
-
2024 - 2027 Seoul, Korea
Ph.D.
Soongsil University
Intelligent Semiconductors
- Advisor: Prof. Seong-heum Kim (Visual Intelligence and Platform Lab)
- Research on efficient deep learning, model pruning, and large vision-language models
-
2022 - 2024 Seoul, Korea
M.S.
Soongsil University
Intelligent Systems
- Thesis on lightweight deep learning for room layout estimation from single panoramic images
- Member of Visual Intelligence and Platform (VIP) Lab
-
2018 - 2022 Seoul, Korea
B.S.
Soongsil University
AI Convergence
Publications
-
2026 One-cycle Structured Pruning with Stability Driven Structure Search
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)
An efficient one-cycle structured pruning framework that integrates pre-training, pruning, and fine-tuning into a single training cycle, achieving state-of-the-art accuracy with minimal training time on CIFAR-10/100, ImageNet, and VGGNet/ResNet/MobileNet/ViT.
-
2024 A Study of Structured Pruning for Hybrid Neural Networks
24th International Conference on Control, Automation and Systems (ICCAS)
Structured pruning for CNN–transformer hybrid architectures with automatic selection of filter pruning criteria and layer-wise pruning rates.
-
2022 A Survey on Efficient Convolutional Neural Networks and Hardware Acceleration
Electronics (MDPI), 11(6), 945
A comprehensive survey on quantized/binarized models, optimized architectures, and resource-constrained systems for efficient deep learning.
-
2022 Lightweight Room Layout Estimation using a Single Panoramic Image
22nd International Conference on Control, Automation and Systems (ICCAS)
Lightweight HorizonNet variant using MnasNet and GRU, reducing parameters to roughly half while maintaining competitive performance.
-
2022 Lightweight Deep Learning for Room Layout Estimation with a Single Panoramic Image
Journal of Institute of Control, Robotics and Systems (KCI), 28(10)
KCI journal version of the lightweight room layout estimation work; validated on Stanford2D3D, PanoContext, and real-world panorama images.
Awards
-
2025 부총리 겸 과학기술정보통신부 장관상 — 시멘틱 세그멘테이션
과학기술정보통신부 (주관 ETRI / IITP / KADIF)
2025 자율주행 인공지능 챌린지: 주행 환경에서의 카메라 기반 픽셀 단위 의미론적 분할.
-
2025 Track 1, 4th Place — Image Classification for Different Lighting Conditions and Styles
IEEE Low-Power Computer Vision Challenge (CVPR 2025 Workshop)
Knowledge-distillation-based lightweight classifier for images under diverse lighting and style conditions.
-
2024 최우수상 (IITP 원장상) — 차량용 객체복합 상태 인식
과학기술정보통신부 (주관 ETRI / IITP / KADIF / 자율주행 DNA 기술포럼)
제1회 자율주행 인공지능 챌린지 — 차량/버스 Agent, 의미론적 Location, 후미등 State 인식 및 Instance Segmentation.
-
2021 우수논문상 — 단일 파노라마 이미지 기반 실내 공간 레이아웃 경량화
숭실대학교 AI융합학부
졸업논문발표회: HorizonNet의 백본을 MnasNet/GRU로 교체하여 약 1/2 파라미터 경량화 달성.
-
2021 장려상 — Face Out
숭실대학교 AI융합학부
AI융합 경진대회 — 얼굴 인식 기반 모자이크 처리 웹 애플리케이션.
-
2021 장려상 — Face Out
숭실대학교
제11회 숭실 캡스톤디자인 경진대회 — 얼굴 인식 기반 모자이크 처리 웹 애플리케이션.
-
2021 우수상 — 코딩 경진대회
숭실대학교 AI모빌리티사업단
숭실 A3+ STAR 코딩 경진대회.
Patents
- Lightweight Deep Learning for Room Layout Estimation with a Single Panoramic Input — 단일 파노라마 입력의 실내 공간 레이아웃 복원 모델 경량화 · Dayoung Kil, Seong-heum Kim · Software Registration No. C-2022-036602 (Republic of Korea, 2022)