MMAI Lab · Ajou University

Multi-Modal Artificial Intelligence Lab

Multi-modal AI research connecting vision, language, and diverse signals for real-world impact.

Our research spans a broad range of machine learning problems, with particular emphasis on computer vision and robot learning. We are also interested in efficient deep learning, especially methods for compressing large language models. Our ultimate goal is to develop multimodal learning frameworks that integrate efficient large-scale vision models and LLMs for robotics, medical data and industrial applications. To this end, publishing our work at top-tier conferences is a key priority for our lab. We welcome inquiries from anyone interested in our research.

Computer VisionRobot LearningEfficient LLMsMedical AIIndustrial AI

MMAI Lab team gathered together for a group photo. — **People**1 Professor, 8 PhD, 6 MS, 10 Interns

Researchers in a meeting room discussing vision AI projects. — **Lab Resources**144 GPUs, 1 Robot

Computer Vision and Learning Algorithms

Visual Recognition

Large-scale Model

Meta Learning

Efficient Learning for LLMs

Model Compression

Efficient Learning

System-level Optimization

Robot Learning

VLA

Efficient Robotics

3D Recognition

Industrial and Medical AI

Battery AI

Fault Detection

Medical AI

Latest News

Latest updates on our research, members, and achievements.

Nov 23, 2026

Paper Accepted

Publication

Re-calibrated Contrastive Loss for Transformation-Aware Prompt Conditioning in Vision-Language Models

British Machine Vision Conference (BMVC) publication update from MMAI Lab.

Seungmin Oh, Seunghun Kang, Jongbin Ryu

Oct 24, 2026

Paper Accepted

Publication

Colla-Q: Toward Collaborative Experts in MoE Quantization via Minimax Precision Balancing

Conference on Empirical Methods in Natural Language Processing (EMNLP) publication update from MMAI Lab.

EunJu Shin, Jongbin Ryu

Oct 24, 2026

Paper Accepted

Publication

Layer-wise Curriculum Learning for Efficient LLM Compression

Conference on Empirical Methods in Natural Language Processing (EMNLP) publication update from MMAI Lab.

Donggeon Lee, Dooyeon Na, Seungmin Oh, Jongbin Ryu

Oct 24, 2026

Paper Accepted

Publication

Train Overcomplete, Deploy Compact: Scaling Recovery Capacity for Structured LLM Pruning

Conference on Empirical Methods in Natural Language Processing (EMNLP) publication update from MMAI Lab.

Seungmin Oh, Donggeon Lee, Jongbin Ryu

Publications

Browse our latest publications and research highlights.

Figure 1 illustrating learnable text prompts and an augmented contrastive space built from image and text encoders over transformed image-text pairs, with adaptive re-calibration reweighting the contrastive loss targets.

Featured publication

Computer Vision and Learning Algorithms

Vision-Language ModelsContrastive LearningPrompt Conditioning

Re-calibrated Contrastive Loss for Transformation-Aware Prompt Conditioning in Vision-Language Models

Seungmin Oh, Seunghun Kang, Jongbin Ryu

British Machine Vision Conference (BMVC) · 2026-11-23

Figure 1 comparing a baseline MoE quantization ensemble that mixes a poorly performing expert into the router's output against Colla-Q, which selects and ensembles collaborative experts for a better aggregated result.

Efficient Learning for LLMs

Mixture-of-ExpertEnsembleQuantization

Colla-Q: Toward Collaborative Experts in MoE Quantization via Minimax Precision Balancing

Conference on Empirical Methods in Natural Language Processing (EMNLP) · 2026-10-24

Figure 1 comparing a baseline that optimizes all student layers at once, causing cumulative error to grow with depth, against the proposed easy-to-hard layer-wise curriculum that passes curriculum features between layers to keep cumulative error low.

Efficient Learning for LLMs

Curriculum LearningLLM CompressionHardware Acceleration

Layer-wise Curriculum Learning for Efficient LLM Compression

Conference on Empirical Methods in Natural Language Processing (EMNLP) · 2026-10-24

Figure 1 comparing LoRA, which keeps pruned layers frozen with low-rank adapters, against the proposed OverRep method, which trains an overcomplete parameterization that is re-parameterized into a compact form at deployment.

Efficient Learning for LLMs

Contact

We welcome inquiries from students, collaborators, and anyone interested in our lab. Please feel free to contact us using the contact information below. We work with many academic and industrial partners on practical learning algorithms and industrial AI applications

Address

Department of Artificial Intelligence, Ajou University

Padal Hall, Room 624,
Department of Artificial Intelligence, Ajou University,
206 World cup-ro, Yeongtong-gu,
Suwon 16499, Korea

General inquiries

jongbinryu@ajou.ac.kr

For collaboration, admissions, and project inquiries, please include your affiliation and topic.

More Information

Multi-Modal Artificial Intelligence Lab

Research Areas

Computer Vision and Learning Algorithms

Efficient Learning for LLMs

Robot Learning

Industrial and Medical AI

Latest News

Publications

Re-calibrated Contrastive Loss for Transformation-Aware Prompt Conditioning in Vision-Language Models

People

Jongbin Ryu

Research interests

Sungeun Kim

Research interests

Eunju Shin

Research interests

HoHyeon Han

Research interests

Contact