IK Inha Kang Deep Learning Researcher

KAIST AI · Vision-Language Models · Reliable Multimodal Learning

AI for the real world.

I am a Ph.D. student at KAIST AI researching vision-language models, compositional understanding, explainability, data-centric AI, and real-world multimodal systems. My work focuses on making AI models more trustworthy, robust, and practical.

View publications Download CV Google Scholar
  • Computer Vision
  • Vision-Language Models
  • Representation Learning
  • Multimodal Learning
11 Papers and preprints
7 First or co-first author works
2× MICCAI MOOD challenge winner
1 Patent

About

Research centered on reliability, reasoning, and deployment.

My work spans vision-language understanding, multimodal reasoning, robust perception, 3D and spatial intelligence, medical imaging, and data-centric approaches that improve AI reliability in practice.

Profile

I am currently pursuing a Ph.D. at KAIST AI after completing an M.S. in Computer Graphics and Visualization at KAIST School of Computing. My research background includes vision-language models, multimodal learning, 3D medical imaging, AR/VR, autonomous driving, and real-world applied AI systems.

Before returning to academia, I worked as a deep learning researcher at FaindersAI, where I built data-centric pipelines, retraining-free model update strategies, multimodal fusion methods, and real-time analytics systems for industry deployment.

01

Compositional VLM Understanding

Studying how multimodal models understand negation, compositional language, and structured reasoning.

02

Explainability and Data-centric AI

Improving trustworthiness by fixing data and reasoning failures, not just scaling model size.

03

3D and Spatial Intelligence

Bridging 2D and 3D understanding for zero-shot spatial reasoning and richer multimodal perception.

04

Real-world AI Systems

Building robust systems for medical AI, forecasting, autonomous driving, and retail analytics.

Selected work

Publication cards designed for fast scanning.

See full publication list
Blind to Position
under-review Vision-Language

Blind to Position, Biased in Language: Failure of Vision-Language Models on Spatial Reasoning

Na-min An*, Inha Kang*, Minhyun Lee, H Shim.

Investigates why VLMs fail at spatial reasoning, revealing positional blindness and linguistic bias, and proposes a diagnostic benchmark with a training-free intervention.

Spatial Reasoning Positional Bias Training-Free
Read paper
What Not to Detect
ICLR 2026 Vision-Language

What “Not” to Detect: Negation-Aware VLMs via Structured Reasoning and Token Merging

Inha Kang, Y Lim, S Lee, J Choi, J Choe, H Shim

Tackles the affirmative bias of VLMs when they encounter negation, combining a token-merging module with a reasoning-aware data pipeline.

Negation Understanding Reasoning Data-centric AI
Read paper
Air Quality Forecasting
CVPR 2026 Forecasting

Real-Time Long Horizon Air Quality Forecasting via Group-Relative Policy Optimization

Inha Kang, E Kim, W Ryu, J Shin, S Yu, YH Kang, S Jeong, E Kim, S Kim, H Shim

Uses GRPO with asymmetric rewards to reduce false alarms and enable reliable five-day East Asia air quality forecasts.

GRPO Foundation Model Fine-Tuning Public Health
Read paper
3D-Aware VLMs
EMNLP 2025 3D VLM

3D-Aware Vision-Language Models Fine-Tuning with Geometric Distillation

S Lee*, J Choi*, Inha Kang, J Kim, J Park, H Shim

Transfers structural knowledge from 3D models into 2D VLMs to improve zero-shot 3D understanding without massive 3D datasets.

Knowledge Distillation 3D Spatial Understanding Zero-Shot
Read paper
Medical Image Anomaly Detection
MICCAI 2022 Medical AI

Joint Embedding of 2D and 3D Networks for Medical Image Anomaly Detection

Inha Kang, J Park

Winning solution for the MICCAI MOOD Challenge, combining local 2D detail and global 3D context for robust anomaly detection.

Challenge Winner 2D/3D Fusion OOD Detection
Read paper

Education

Academic journey

Aug 2024 – Present Ph.D.

KAIST AI

Computer Vision and Machine Learning Lab · Advisor: Prof. Hyunjung Shim

Research focus: Vision-Language Models, Multimodal Learning · GPA 4.3 / 4.3

Aug 2020 – Aug 2022 M.S.

KAIST School of Computing

Computer Graphics and Visualization Lab · Advisor: Prof. Jinah Park

Research focus: 3D Medical Imaging, AR/VR · GPA 3.88 / 4.3 · MICCAI MOOD Challenge 1st Place in 2021 and 2022

2017 Visit

University of California, Berkeley

Computer Science

Mar 2016 – Aug 2020 B.S.

GIST

Electronic Engineering and Computer Science

Graduated salutatorian · Best Thesis Award · Minor in Business Administration and Sociology

Experience

Research and industry

2022 – 2024 Industry

FaindersAI · Deep Learning Researcher

Seoul, Korea

  • Built a data engine to generate high-quality segmentation data using a feature-based VLM filtering pipeline.
  • Studied retraining-free model update strategies and multimodal fusion across diverse sensor inputs.
  • Engineered a real-time retail analytics system that achieved 2× competitor accuracy.
2019 – 2020 Research

Machine Learning and Vision Lab · Research Intern

Gwangju, Korea

  • Designed a scheduled training method for robust long-tailed object detection in autonomous driving.
2017 – 2018 Research

Aerospace Propulsion Lab · Research Intern

Gwangju, Korea

  • Studied parallel computing and GPU acceleration for high-performance workloads.

Projects

Applied systems across medical AI, XR, and human-centered design.

View all projects
Medical Imaging

PM2.5 Interaction Visualization

3D label-free visualization of fine dust interaction with cells using optical diffraction tomography.

XR / AR / VR

SketchScope

Interactive virtual science museum platform with sketch-based volume exploration, VR interaction, and HoloLens 2 support.

Project page
Human-Centered AI

ForMixMatch

Virtual fashion coordination service exploring image-based try-on and layout design for confident online shopping decisions.

Project deck

Contact

Open to research collaboration and applied AI projects.

I am especially interested in reliable multimodal learning, vision-language reasoning, data-centric AI, and practical systems where robustness matters.

Email me LinkedIn GitHub