Jaemin Choi
M.S. student in Electronic Engineering
Sogang University, Seoul, South Korea
I'm an M.S. student at Sogang University, advised by Prof. Suk-Ju Kang in the VIDS Lab. I work on efficient inference for foundation models in computer vision: making them lighter and faster to run while keeping their performance.
My research covers post-training quantization, KV cache compression, diffusion caching, and long-term memory for large vision models.
- Post-Training Quantization
- KV Cache Compression
- Diffusion Caching
- Long-Term Memory
Publications
* Equal contribution † Corresponding author
Top-Tier Conference
-
Conference on Neural Information Processing Systems (NeurIPS 2026) Post-Training Quantization with Gradient-Projected Fisher Approximation for Vision Transformers
-
Conference on Neural Information Processing Systems (NeurIPS 2026) SPA-Q: Structure-Preserving Adaptive Post-Training Quantization for Monocular Depth Estimation
-
Conference on Neural Information Processing Systems (NeurIPS 2025) QSCA: Quantization with Self-Compensating Auxiliary for Monocular Depth Estimation
Workshop
-
International Conference on Machine Learning Workshop, AdaptFM (ICMLw 2026) Implicit Off-Diagonal Curvature Modeling via Gradient Projection for Post-Training Quantization of Vision Transformers
-
International Conference on Machine Learning Workshop, AdaptFM (ICMLw 2026) Structure-Preserving Adaptive Post-Training Quantization for Monocular Depth Estimation
-
International Conference on Machine Learning Workshop, AdaptFM (ICMLw 2026) Structural Outlier-Aware Post-Training Quantization for Monocular Depth Estimation
-
International Conference on Machine Learning Workshop, AdaptFM (ICMLw 2026) Reducing Attention Distribution Error with Unified Tail Aggregation for Sparse Attention
Conference
-
Society for Information Display, Display Week (SID 2026) Enhancing Perceptual Quality on High-Resolution Displays: A Unified Deep Model for Super-Resolution and Deblurring
-
International Technical Conference on Circuits/Systems, Computers, and Communications (ITC-CSCC 2025) Image Denoising Meets Quantization: Exploring the Effects of Post-Training Quantization
Under Review
-
Future-Query-Guided KV Cache Compression for Visual Autoregressive Modeling
-
SparSTAR: Sparse Attention for Spatiotemporal Autoregressive Video Synthesis
-
ResceneKV: Reconstruction-Error Selective KV Cache Eviction across Nested Scales for Visual Autoregressive Modeling
-
Towards a Low-Delta Oracle Policy for Improved Upper Bound in Sparse Attention
-
OpenAttention: One Percent is ENough with Variance-Aware Tail Aggregation for Sparse Attention
-
NBAFE: NPU-friendly Blur-Aware Frequency Enhancement for Single Image Defocus Deblurring
Research Experience
-
Mar 2025 – Present
Samsung Display Research collaboration with the Computer-Aided Engineering team
Developing AI technology for image quality restoration.
Experience
Technical Academic Activities
-
2026
Reviewer, ICLR 2027 International Conference on Learning Representations