Finding signal in the noise.
鼓動・遊び・対比
I build machine-learning systems that pick out faint signals in noisy data. Most of the work is the engineering that makes them trustworthy: validating them with clinicians against clinical reference standards, and keeping them reliable once they are in clinical use.
Selected work仕事
A few things I’ve built: clinical signal analysis, learned reconstruction, and models you can interrogate.
Approach手法
Four threads run through my work, whether the signal comes from a heart or a telescope.
Signal detection in noise信号検出
Arrhythmia and anomaly detection, denoising, and time-frequency analysis on noisy physiological signals. Lomb-Scargle periodograms, wavelets, and matched filtering in NumPy and SciPy, feeding LightGBM models.
Deep learning at scale大規模学習
Training models on millions of records in PyTorch and JAX, across multi-GPU and HPC clusters with CUDA. Reproducible data pipelines and the ordinary engineering that keeps a model dependable long after the prototype.
Explainable & trustworthy AI説明可能性
Model interpretability with SmoothGrad, Grad-CAM, and integrated gradients, packaged in my open-source NeuroExplain library. Calibration and evaluation in scikit-learn, so clinicians can check why a model made its call.
Research engineering & systems研究基盤
Reproducible, well-tested research software in Python, with Rust and C++ where performance matters. Docker, CI/CD, and pytest on long-lived codebases, with recent experiments moving the training stack onto Kubernetes.
Recognition評価
Peer-reviewed publications, competitive funding, and selective research programmes.
Biological Psychiatry: Global Open Science
Peer-reviewed journal · µ-opioid modulation of sensorimotor connectivity (co-author)
PhD, Machine Learning & Neuroimaging
King’s College London · explainable AI for brain connectivity
OHBM: Explainable deep learning for subtyping
SmoothGrad attribution · Organization for Human Brain Mapping
£100k UKRI innovation funding
Tycho MedLink · UCL & Cambridge Judge accelerators
OHBM: Neurodevelopmental phenotypes via GNNs
Neonatal brain connectivity · graph neural networks
Google Summer of Code
Infant eye-tracking API · mentored by McGill Ophthalmology
ISMRM: 3D-VNN coronary MR angiography
Learned reconstruction · Int’l Society for Magnetic Resonance in Medicine
Contact連絡
Happy to talk about interesting problems in signals and machine learning. Email is the fastest way to reach me.
Open to
- Research and engineering roles in signals and machine learning
- Collaborations on time-series and detection problems
- Speaking, peer review, and technical writing