Research directions
Projects
My research builds on industrial and nonlinear control and is expanding toward large language models, multimodal reasoning, and generative systems. These projects explore how learned models can support perception, modeling, and decision making.
Public Research
Hierarchical and Step-Layer-Wise Tuning of Attention Specialty for Multi-Instance Synthesis in Diffusion Transformers
Generating several distinct objects in one image requires control over their placement and individual attributes. This work studies how different token types are represented across diffusion-transformer layers and introduces training-free attention specialty tuning to guide multi-instance synthesis across layers and denoising steps.
Current Research
The following projects are in progress.
Generative Models & Multimodal Reasoning
Training-Free Consistent Synthesis and Editing in Diffusion Transformers
Exploring training-free methods to preserve identity and structure during image generation and editing.
Learnable Persistent Memory Router for Narrative Long Multi-Shot Video Generation
Investigating persistent memory mechanisms to maintain coherence across shots in long-form video generation.
VLM-Based Multi-Agent Systems for Long and Complex Video Understanding
Exploring how cooperating vision-language models can reason over long videos and connect information across events.
Learning for Energy & Dynamical Systems
Conditional Flow Matching for Multivariate Energy Profile Imputation
Investigating conditional generative models for filling missing values in multivariate energy profiles.
Optimal Control of Infinite-Dimensional Systems via Neural Operators
Exploring neural-operator methods for modeling and controlling systems whose dynamics vary over space and time.