Welcome to ODAI Lab.

At ODAI@UOS, our research delves into cutting-edge on-device AI technologies, aiming to bring advanced intelligence directly to edge devices. We primarily focus on three critical areas: Lightweight (경량) AI, Secure (보안) AI, and Physical (물리) AI.

1. Lightweight AI: We develop highly efficient AI models for resource-constrained edge devices. Our research maximizes the energy efficiency of Spiking Neural Networks (SNNs) and Large Multi-modal Models (LMMs). We also advance structural optimizations, such as extremely low-bit (e.g., 4-bit) quantization, for Open-Vocabulary Object Detection (OVOD). 

2. Secure AI: We enhance the robustness of AI models against adversarial attacks and data breaches. Our work investigates vulnerabilities targeting the temporal dynamics of Multi-Object Trackers (MOT) and SNNs, and develops fundamental security frameworks, including TEE-shielded DNN isolation and defenses against embedding inversion in Multi-Agent RAG systems. 

3. Physical AI: We build intelligent systems guaranteeing strict real-time performance in dynamic environments, such as autonomous vehicles and LEO satellites. To optimize the latency-accuracy trade-off under limited resources, we design dynamic scheduling frameworks integrated with Coarse-to-Fine Transformers (CF-DETR) and real-time MOT systems. 

News

• [February 2026] A paper entitled "Adaptive Spiking Neural Networks for Real-Time Multi-Object Detection Tasks" has been accepted to DAC'26 (BK21 IF3).

• [November 2025] A paper entitled "Timestep-Compressed Attack on Spiking Neural Networks through Timestep-Level Backpropagation" has been accepted to AAAI'26 (BK21 IF4).

• [September 2025] NRF 미래도전연구지원사업 과제선정, "초저전력 뉴로모픽 AI 반도체 기반 실시간 LMM 플렛폼 연구", 2025.9-2030.8, 연2억(총10억)원, 연구책임자: 백형부

• [September 2025] 과학기술정보통신부 / NRF 주관 2025 K-DATA Science 해커톤 창의상 수상, 학부연구생팀: 강동훈, 한지인, 이현승

• [September 2025] A paper entitled "ARES: Adaptive Robust Object Detection Framework for Enhancing Real-time Performance in Autonomous Vehicle Systems" has been accepted to JSA (JCR Q1).

• [August 2025] A paper entitled "CF-DETR: Coarse-to-Fine Transformer for Real-Time Object Detection" has been accepted to RTSS'25 (BK21 IF4).

• [May 2025] A paper entitled "BankTweak: Adversarial Attack against Multi-Object Trackers by Manipulating Feature Banks" has been accepted to IJCAI'25 (BK21 IF4).

• [January 2025] A paper entitled "Real-Time Scheduling for Multi-Object Tracking Tasks in Regions with Different Criticalities" has been accepted to JSA (JCR Q1).

• [August 2024] Three papers have been accepted to ICPR'24 (BK21 IF1).

• [July 2024] A paper entitled "Batch-MOT: Batch-Enabled Real-Time Scheduling for Multi-Object Tracking Tasks" has been accepted to EMSOFT'24 (BK21 IF2) and TCAD.

• [March 2024] On-Device Artificial Intelligence (ODAI) laboratory has been launched at the University of Seoul (UOS).