Cedric Lim — PhD Candidate, Materials Science and Engineering @ Stanford University ·
Reconstructing the invisible, one tomogram at a time.
I work at the intersection of materials science, tomographic imaging, and deep learning — using AI to reconstruct 3D structure from limited measurements.
// Currently
Research
Finishing a paper on physics-informed neural networks for atom-probe tomography.
Reading
Deep Learning for the Sciences (Davies et al.)
Building
An open-source pipeline for sparse-view electron tomography.
Location
Stanford, CA — back from a research stay in Zürich.
Selected work — 2023 / 2025
All projects →Physics-informed Tomography Networks
A neural reconstruction framework that bakes mass-conservation and geometric priors into the loss for sparse-view electron tomography.
PyTorchJAXCUDA
AtomVis
A WebGL viewer for atom-probe datasets — handles 100M+ ions interactively in the browser with progressive LOD.
WebGLReactRust/WASM
DiffRec
Diffusion priors for ill-posed inverse problems in 4D-STEM. Recovers strain fields from 5× fewer diffraction patterns.
PyTorchDiffusers
Labnotes
A tiny CLI that turns markdown lab notebooks into a searchable static site. Built in a weekend; quietly used by my whole group.
GoLit
Writing
All posts →How I structure a PhD-scale data pipeline
Three years in, I've finally settled on a setup that survives both reviewer #2 and a 2 a.m. cluster crash.
Physics priors are a free lunch (when you season them right)
Why baking conservation laws into the loss almost always beats post-hoc constraints — and the one case where it doesn't.
Publications
Full CV →Physics-informed neural networks for sparse-view atom probe tomography reconstruction
Diffusion priors for inverse problems in 4D scanning transmission electron microscopy
AtomVis: interactive visualization of large-scale atom probe data on the web
Self-supervised denoising for low-dose electron tomography