Dayoung Kil

Ph.D. Student

Graduate School of Soongsil University

Research on hardware-efficient AI & large vision-language models.

Seoul, Republic of Korea

dayoungkil (at) soongsil.ac.kr

I am Dayoung Kil, a Ph.D. Student at Soongsil University’s Graduate School, affiliated with the Visual Intelligence and Platform (VIP) Lab under the supervision of Prof. Seong-heum Kim.

My research lies at the intersection of computer vision and efficient deep learning, with a focus on making large-scale vision and multimodal systems tractable under realistic compute budgets. I am currently working on token pruning and efficient inference for Large Vision-Language Models (LVLMs) and multimodal LLMs — building on earlier work in structured network pruning and lightweight architectures for visual perception.

Broadly, I am driven by questions at the accuracy–compute frontier: how far can we push the efficiency of modern vision-language systems without trading away their capability?

Research Interests

Large Vision-Language Models (LVLMs) & Multimodal LLMs
Token pruning and efficient inference for transformers
Structured network pruning & model compression
Hardware-efficient deep learning
Computer vision

news

May 18, 2026	Team SSUPER_POWER finished 2nd place in Track 3 (AI Generated Images Detection) of the 2026 IEEE Low-Power Computer Vision Challenge, held at the ECV Workshop at CVPR 2026, Denver, CO — sponsored by Qualcomm. (results)
Jan 15, 2026	The paper One-cycle Structured Pruning with Stability Driven Structure Search has been accepted to WACV 2026.
Nov 14, 2025	Received the 부총리 겸 과학기술정보통신부 장관상 at the 2025 자율주행 인공지능 챌린지 (Semantic Segmentation track).

selected publications

WACV
One-cycle Structured Pruning with Stability Driven Structure Search

Deepak Ghimire, Dayoung Kil, Seonghwan Jeong, and 2 more authors

IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2026

Abs arXiv Bib HTML Code

Existing structured pruning typically involves multi-stage training procedures that often demand heavy computation. We propose an efficient framework for one-cycle structured pruning that integrates pre-training, pruning, and fine-tuning into a single training cycle. The core idea is to search for the optimal sub-network during the early stages of network training, guided by norm-based group saliency criteria and structured sparsity regularization. We introduce a novel pruning indicator that determines the stable pruning epoch by assessing the similarity between evolving pruning sub-networks across consecutive training epochs. Extensive experiments on CIFAR-10/100 and ImageNet using VGGNet, ResNet, MobileNet, and ViT architectures demonstrate state-of-the-art accuracy while being one of the most efficient pruning frameworks in terms of training time.
@article{kil2025ocspruner, title = {One-cycle Structured Pruning with Stability Driven Structure Search}, author = {Ghimire, Deepak and Kil, Dayoung and Jeong, Seonghwan and Park, Jaesik and Kim, Seong-heum}, journal = {IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)}, year = {2026}, }
Electronics
A Survey on Efficient Convolutional Neural Networks and Hardware Acceleration

Deepak Ghimire, Dayoung Kil, and Seong-heum Kim

Electronics, 2022

Abs DOI Bib HTML

Over the past decade, deep-learning-based representations have demonstrated remarkable performance in academia and industry. In this review, to improve the efficiency of deep learning research, we focus on three aspects: quantized/binarized models, optimized architectures, and resource-constrained systems. Recent advances in light-weight deep learning models and network architecture search (NAS) algorithms are reviewed, starting with simplified layers and efficient convolution and including new architectural design and optimization.
@article{ghimire2022survey, title = {A Survey on Efficient Convolutional Neural Networks and Hardware Acceleration}, author = {Ghimire, Deepak and Kil, Dayoung and Kim, Seong-heum}, journal = {Electronics}, volume = {11}, number = {6}, pages = {945}, year = {2022}, publisher = {MDPI}, doi = {10.3390/electronics11060945}, }