I work on the part of machine learning that has to decide what to measure next — and on making that decision cheap enough to run on real hardware, in real time.
PhD in statistics from NTNU (2023): value of information and informative path planning for autonomous underwater vehicles. Gaussian random fields, decision-theoretic acquisition criteria, and a long-horizon planner we ran onboard for 2.5 hours in a Norwegian fjord.
Since 2023 at Sandvik Coromant in Trondheim, building real-time anomaly detection over multi-sensor industrial streams — LSTM autoencoders under 16 ms inference, CUDA training pipelines, ONNX Runtime on embedded targets, in production.
Most of the two halves rarely meet. The interesting problems live in between.
- Real-time ML on constrained hardware — PyTorch, CUDA, ONNX Runtime, embedded deployment
- Decision-theoretic sensing — Gaussian processes and random fields, value of information, informative path planning
- HPC and distributed computing — MPI, OpenMP, Slurm, GPU clusters
- The rest of the stack — Python, C++, C#/.NET, Blazor, TypeScript, Azure CI/CD
Most of my day-to-day work is proprietary. What's public here is the research code and the writing.
- RRT*-enhanced long-horizon path planning for AUV adaptive sampling using a cost valley — Knowledge-Based Systems 315 (2025) · paper · code
- 3D adaptive AUV sampling for the classification of water masses — IEEE Journal of Oceanic Engineering 48 (2023)
- Efficient 3D real-time adaptive AUV sampling of a river plume front — Frontiers in Marine Science 10 (2024)
- Using expected improvement of gradients for robotic exploration of ocean salinity fronts — Environmetrics 36 (2025) · paper
- Long-horizon informative path planning with obstacles and time constraints — IFAC-PapersOnLine 55 (2022)
Full list, with PDFs → yaolinge.com