Chashi Mahiul

Motivation to pursue a graduate degree

I was always fascinated by how AI systems actually work under the hood, not just how to use them. The more I learned, the more I realized how many fundamental questions are still unanswered, and that pulled me deeper. I chose FSU because of Dr. Xiuwen Liu, whose work on representation learning and neural network geometry aligned perfectly with what I wanted to investigate. The department gave me the freedom and resources to pursue research that is both theoretically grounded and practically meaningful.

Importance and/or impact of research and work

I study why foundation models like vision transformers fail in ways that are invisible to humans but catastrophically exploitable by machines. Specifically I look at how the geometry of embedding spaces creates vulnerabilities where visually identical images can have completely different representations, and visually different images can share the same one. This matters because these models are being deployed at massive scale in recommendation systems, agentic pipelines, and medical AI, and nobody has systematically mapped out where they break. Understanding the failure modes is the first step toward building AI systems that are actually trustworthy.

Career aspirations

My goal is to build AI systems that are not just powerful but genuinely reliable, either leading research at an industry lab or as a professor pushing the theoretical foundations forward. I want the work I am doing now on representation vulnerability and geometric deep learning to eventually shape how the next generation of foundation models are designed and evaluated.

Advice for anyone considering graduate school

Only come if you are genuinely obsessed with a question, not just looking for a credential, because the credential alone is not worth the hard parts. The research will frustrate you, humble you, and occasionally thrill you, and you need to be chasing the thrill to survive the frustration. Find an advisor whose thinking you respect and whose working style you can grow in, because that relationship will define your entire experience.

Accomplishments during graduate career

I am most proud of publishing at IROS 2024 and WCCI 2026, both A rank conferences. Getting work accepted at top venues meant that the questions I was asking and the methods I was developing were being taken seriously by the broader research community. What makes these especially meaningful is that both papers came from genuinely curiosity driven work.