Evolution of Semantics in Diffusion Models
Ongoing research analyzing semantic feature evolution during diffusion denoising.
Research Overview
Current Ongoing Research
I am currently investigating the internal dynamics of diffusion models, specifically analyzing how semantic features evolve throughout the iterative denoising process.
This work aims to provide deeper interpretability into how diffusion models “understand” and construct complex visual concepts from noise, potentially leading to more controllable and efficient generation methods.
Tech Stack
- Deep Learning Interpretability
- Diffusion Models
- Python / PyTorch