Role: Research Intern (Lead Author) | Tech: PyTorch, EEG, CLIP, Transformers
Proposed a novel dual-stream EEG encoder inspired by the brain's ventral ("what") and dorsal ("where") visual pathways. Aligned EEG signals with hierarchical CLIP embeddings to significantly improve zero-shot image retrieval accuracy on the THINGS-EEG2 dataset.
Presented at KSMBE 2025 (Poster)
Targeting IEEE BCI Winter 2026
Role: Team Member | Rank: 18 / 1,183 (Top 1.5%)
Achieved Rank 18th out of 1,183 teams (Top 1.5%) in the "Cross-Task to Cross-Subject EEG Decoding" challenge. Utilized the HBN-EEG dataset (3,000+ subjects) to develop models for Cross-Task Transfer Learning and Externalizing Factor Prediction. Achieved a final overall score of 0.98008 on Codabench.
Role: Team Leader | Award: Excellence Award (Graduation Project)
Developed an end-to-end pipeline that generates interactive VR environments from a single text prompt. Integrated DreamScene360 for panoramic 3D Gaussian Splatting and Hunyuan3D-2.0 for interactive asset generation, automating Unity scene composition.
Submitted to HCI Korea 2026
Role: Team Leader | Award: First Prize 🏆
Built a custom computer vision model from scratch to classify solar features and perform object detection on solar images. Achieved robust performance in detecting specific solar phenomena and identifying their precise locations.
Role: Team Leader
Engineered a BCI system analyzing P300 ERP signals to decode clothing preferences and generate virtual try-on images using IDM-VTON.
Role: Solo Project
Developed a real-time ECG monitoring system on STM32F4xx, processing 100Hz signals to control actuators based on heart rate.
Role: Team Member (Top 8%)
Developed an ensemble model (ViT, Swin) to predict gene expression profiles from pathology images.
Role: Team Member (Awarded)
Fine-tuned DenseNet-121 to classify wisdom tooth extraction risk, demonstrating transfer learning skills.
Role: AI Design
Designed an AI service for objective peer evaluation using quantitative metrics from team interactions.
Role: Developer
Implemented a pipeline using prompt engineering, segmentation, and diffusion models to synthesize clothing.