Research statement
Intelligence should adapt, reason, and earn trust.
I am interested in artificial intelligence that learns from multiple signals, interacts with its environment, and changes its behaviour without discarding the useful structure it has already acquired.
Learning from more than one view
Human intelligence is shaped by the interaction of language, vision, sound, memory, and action. Multimodal models offer a path toward systems that form similarly connected representations. My work asks how those models can reason over diverse evidence, adapt efficiently to unfamiliar settings, and remain dependable when the world differs from their training data.
Changing models carefully
Agents must also change over time. They may need to absorb new capabilities, correct a behaviour, or respond to new environments without losing what already works. I study this challenge through trustworthy AI agents, parameter-efficient adaptation, machine unlearning, multimodal reasoning, and verification.
Research through clear questions
Research rarely moves in a straight line. Negative results are useful when they sharpen the question and reveal which assumptions were doing hidden work. I try to make each project legible: state the problem plainly, design an experiment that can contradict the hypothesis, and follow the evidence.