Projects

• The Hidden Evolution of Disguised Visual Context inside the VLM
  A study of how different vision integration paradigms behave in VLMs, and how each one transforms the visual representation inside the LLM.
• CoLA: Cross-Modal Low-rank Adaptation for Multimodal Downstream Tasks
  A PEFT framework that extends LoRA with a dedicated inter-modal adaptation pathway, for adapting dual-stream unimodal encoders to multimodal tasks.