I am a postdoctoral researcher working on reasoning in diffusion language models and safety in vision-language models. My research aims to understand and improve the generative capabilities of diffusion-based architectures on language, multimodal data, and to develop robust safety mechanisms for multimodal AI systems.
Enabling reasoning capabilities in diffusion-based language models, including distillation of AR reasoning traces and investigating semantic superposition in continuous token spaces.
Concept erasure in text-to-image diffusion models, safety classification for generated content, and understanding differential erasure difficulty across concept types.
Diffusion processes, masked diffusion, flow matching, and their mathematical foundations for both language and vision domains.
RL-based fine-tuning for diffusion models, and developing evaluation frameworks for safety and robustness in multimodal systems.