Efficient Representations
Linear AlgebraEfficiencyEmbedded Devices
Exploiting Boosting in Hyperdimensional Computing for Enhanced Reliability in Healthcare
Publications
A connected view of my research trajectory, followed by accepted papers and arXiv preprints.
Research Trajectory
My research has evolved from efficient representation learning, to multimodal modeling and uncertainty, then to geometric interpretation of representation spaces, multimodal agentic reasoning, and internal analysis for controllable agentic systems.
01
→
02
→
Multimodal Learning
MultimodalCross-Modal AlignmentEvent Streams
Cross-Modal Event Encoder: Bridging Image-Text Knowledge to Event Streams
03
→
Probabilistic Fusion
ProbabilisticBayesian UncertaintyRobustness
Uncertainty-Weighted Image-Event Multimodal Fusion for Video Anomaly Detection
04
→
Geometric Interpretation
Linear AlgebraGeometryVector Space
Understanding the Visual Projection Space of Multimodal LLMs
05
→
Multimodal Agents
MultimodalAgentic Expert OrchestrationVisual Reasoning
Draft and Refine with Visual Experts
06
→
Hallucination Analysis
Linear AlgebraGeometryFlow Signatures
Internal Flow Signatures for Self-Checking and Refinement in LLMs
07
Controllable Agents
AI AgentControlInterpretabilityState-Centric Reasoning
State-Centric Decision Process
Research Blueprint
A conceptual map of how research themes connect across papers. Time flows from top to bottom.
ThemesPapers
Lime: efficiencyEmerald: linear algebraBlue: multimodalPurple: probabilityTeal: hallucinationRose: agenticAmber: prior paper
Accepted Publications
[5]Draft and Refine with Visual Experts
SungHeon Jeong, Ryozo Masukawa, Jihong Park, Sanggeon Yun, Wenjun Huang, Hanning Chen, Mahdi Imani, Mohsen Imani
An agent framework that improves multimodal reasoning by measuring visual reliance and refining responses with feedback from visual experts.
[3]Understanding the Visual Projection Space of Multimodal LLMs
SungHeon Jeong, Yoojeong Song, Hyungjoon Kim
A geometric probing study of the projected visual token in multimodal LLMs, analyzing latent-token alignment, intrinsic dimensionality, and perturbation sensitivity.
[2]Cross-Modal Event Encoder: Bridging Image-Text Knowledge to Event Streams
SungHeon Jeong, Hanning Chen, Sanggeon Yun, Suhyeon Cho, Wenjun Huang, Xiangjian Liu, Mohsen Imani
A cross-modal event encoder that adapts CLIP's image-text representation space to event streams while preserving zero-shot learning and text alignment.
[1]Exploiting Boosting in Hyperdimensional Computing for Enhanced Reliability in Healthcare
SungHeon Jeong, Hamza Errahmouni Barkam, Sanggeon Yun, Yeseong Kim, Shaahin Angizi, Mohsen Imani
A hyperdimensional computing framework that applies boosting to improve reliability and robustness in healthcare-oriented learning tasks.
arXiv Preprints
[7]State-Centric Decision Process
SungHeon Jeong, Ryozo Masukawa, Sanggeon Yun, Mahdi Imani, Mohsen Imani
A state-centric framework for agent decision-making that represents reasoning trajectories through certified state transitions and supports analysis such as credit assignment, failure localization, and modular operator replacement.
[6]Internal Flow Signatures for Self-Checking and Refinement in LLMs
SungHeon Jeong, Sanggeon Yun, Ryozo Masukawa, Wenjun Huang, Hanning Chen, Mohsen Imani
A self-checking and refinement framework that audits internal decision dynamics of LLMs and enables targeted correction without modifying the base model.
In Submission
[4]Uncertainty-Weighted Image-Event Multimodal Fusion for Video Anomaly Detection
SungHeon Jeong, Jihong Park, Mohsen Imani
A multimodal video anomaly detection framework that fuses image and event representations using uncertainty-aware weighting for robust anomaly localization.