WorkPhD Researcher · King's College London · 2020 — 2024
Explainable AI for Neuroimaging説明可能性
Making deep models legible for clinical neuroscience.
My doctoral work built attribution methods — SmoothGrad, Grad-CAM, integrated gradients — to reveal which features drive a model’s predictions from brain-connectivity data, so the outputs could be trusted and interrogated.
What I do
- Adapted gradient-based attribution to graph and imaging models.
- Open-sourced tooling (NeuroExplain).
- Presented at OHBM 2023 and 2024.