STOCHOS is built on probabilistic machine learning for engineering data: small, noisy, expensive. One algorithm, DIM-GP, powers all six methods.
Start with the algorithm
Everything in STOCHOS rests on DIM-GP, a Deep Infinite Mixture of Gaussian Processes: deep learning capacity with Gaussian process uncertainty, in one model built for small, noisy, expensive data.
Read how STOCHOS works: the DIM-GP algorithm →
Click to train a model
This is a live Gaussian process, the building block DIM-GP extends, running in your browser. Click anywhere in the plot to add a measurement. The model refits, and the amber band shows its confidence: tight where data exists, wide where it is guessing.
Six methods, one loop
Each method is a stage of one loop around the model: model, trust, decide, cut cost, explain, expand. Every card opens a plain-language guide showing how STOCHOS runs that stage.
Surrogate Modeling
Model it: predict new variants in seconds from the runs you already have.
Uncertainty Quantification
Trust it: see where predictions are reliable and where more data would help.
Bayesian Optimization
Decide: choose the next best experiment or simulation.
Multi-Fidelity Modeling
Cut the cost: combine cheap screening data with expensive high-quality results.
Sensitivity Analysis
Explain it: identify which inputs actually drive the outcome.
Generative Design
Expand it: explore new geometries beyond fixed variants with GEN-BO.
In a DLR and SGL Carbon project presented at the DPG Annual Conference, DIM-GP replaced the microscale FEM step in a multi-scale fracture framework. Read the project note
PI Probaligence took first prize at the Robert Bosch Foundation Health Award, for work outside engineering. Read the announcement
