Research Interest

My research lies at the intersection of Computer Vision and Natural Language Processing, with a focus on understanding and improving Vision-Language Models (VLMs). I study how these models encode visual and semantic information, how meaningful signals can be extracted from their internal representations (e.g., attention and embeddings), and how their behavior varies across tasks, prompts, and distribution shifts.

During my PhD, I focused on semantic segmentation in data-limited settings, developing methods that learn efficient representations and leverage weak supervision and minimal annotations. Building on this foundation, I continue to explore approaches that reduce reliance on labeled data, including techniques that utilize unlabeled test images through test-time adaptation.

As a Postdoctoral Researcher, my current work centers on analyzing VLM behavior and comparing model attention patterns with human attention, with the broader goal of designing multimodal systems that enable more reliable decision-making in sensitive real-world applications.