Probabilistic AI for physical systems.

STOCHOS learns from your simulation, measurement, and test data, predicts new variants in seconds, and puts a confidence on every output.

Confidence narrows with more data and widens with less.

How it works

From your data to the DIM-GP engine

Your data goes in, the model learns the relationships, and every prediction comes back with a confidence.

Start

Your data

Simulation results, measurements, lab and process data, test records.

STOCHOS

Learns the relationships

Which inputs drive the output, how they interact, and where data is sparse.

What you get

Predictions in seconds

Outcomes for variants and operating points you have not evaluated yet.

Why it matters

Confidence on every output

See where the model is reliable and where another run should come next.

The DIM-GP model

The DIM-GP architecture: a deep neural network and a second network feed learned basis functions into Gaussian process blocks that combine into one probabilistic field prediction.

DIM-GP: deep learning with Bayesian uncertainty

The pattern recognition of deep learning, the uncertainty estimation of Gaussian processes, one model built for small, noisy datasets.

Read the full DIM-GP explanation →

Supported data

Built for real engineering and R&D data

One probabilistic method spans engineering, chemistry, and formulation data that usually needs several separate tools.

A dense grid of glowing amber and grey numbers, a tabular dataset.

Scalar and tabular

Material properties, process parameters, recipes, test results.

An amber signal curve rising over time with a soft uncertainty band, on a faint grid.

Signals and time series

Vibration curves, force-displacement, spectra, process histories.

Amber airflow streamlines over an airfoil with a turbulent wake, a CFD flow-field result.

2D and 3D fields

CFD pressure fields, FEM stress maps, temperature distributions.

A grid of sixteen amber microscopy tiles of different material surface textures.

Images

Microscopy, thermography, surface and defect inspection frames.

A wireframe FEM mesh of a mechanical bracket, glowing amber on black.

Geometries, meshes, and volumes

FEM meshes and graphs, STL geometries, point clouds, and volumetric fields.

A ball-and-stick molecular model in glossy amber on black.

Formulations and chemistry

Compositions, raw materials, concentrations, molecular structure.

Core capabilities

One intelligence layer, six decision tools

Surrogate modelingTrain a fast stand-in for an expensive simulation, experiment, or process. Uncertainty quantificationSee where a prediction is reliable and where more data is needed. Bayesian optimizationFind the best design, recipe, or operating point in fewer runs. Generative designCreate valid geometry variants and search them efficiently with GEN-BO. Sensitivity analysisRank the variables that actually drive the outcome. Multi-fidelity modelingFuse many cheap simulations with few precise tests in one model.

Built as a library

Train and predict with an AI model in 4 lines of code

STOCHOS is a Python library, not a black box. Fit on small, noisy data, then predict with a confidence interval.

surrogate.py

# train an uncertainty-aware surrogate
from stochos.dimgp import dimgp_regr

model = dimgp_regr()
model.fit(X_train, Y_train)          # small, noisy data is fine

# prediction with a confidence interval
Y_pred, Y_lower, Y_upper = model.predict(X_new, CI=0.95)

Real STOCHOS API (dimgp_regr) · Python on PyTorch · Windows and Linux · CPU or GPU

A few lines of code on a normal workstation, not a server: a modern CPU and about 16 GB RAM.

Fits your stack

Sits alongside NumPy, pandas, PyTorch, Jupyter, and your solver scripts.

Automatable and reproducible

Script sweeps, batch jobs, and validation so results are repeatable.

Your data stays local

Runs on your own infrastructure and installs offline. Cloud is optional, never required.

Your AI workflow is just one prompt away

Tell the Workflow Agent what you need and it builds the workflow for you. Plug in the tools you already use, adjust any step, rerun on new data. STOCHOS does the predicting underneath.

Explore STOCHOS Flow →

Give the optimizer a problem and see how close your own best try lands. Take the challenge

See what STOCHOS can do with your data

Bring one use case, one bottleneck, or one dataset. We can outline where STOCHOS fits, which method is relevant, and what a realistic proof of concept could look like.

Request a Demo Explore STOCHOS Flow

Common questions

01What is a surrogate model?

A surrogate model is a data-driven approximation of an expensive simulation or experiment. It learns the relationship between inputs and outputs from data you already have, then predicts results for new variants in seconds instead of rerunning the full solve or test each time.

02What is uncertainty quantification?

Uncertainty quantification reports how confident a model is in each prediction. Instead of one point value, you get a value and a measure of its reliability. This lets engineers trust predictions inside the data range and run a full validation only where confidence is low.

03What algorithm does STOCHOS use?

DIM-GP, a Deep Infinite Mixture of Gaussian Processes, is the core STOCHOS algorithm. It combines neural networks and Gaussian processes into a non-stationary, probabilistic model designed for small, noisy datasets, changing geometries and meshes, and multi-fidelity data, without manual hyperparameter tuning.

04What data types does STOCHOS support?

STOCHOS works with scalar and tabular data, signals and time series, 2D and 3D fields from CFD and FEM, particle and transient simulation results, graphs and meshes, geometries and point clouds, images, and multi-fidelity data. This breadth lets one approach cover engineering, chemistry, and formulation problems.

05What hardware does STOCHOS need?

A normal workstation, not a server. Core modelling, optimization, and sensitivity run on a modern CPU with about 16 GB RAM, and core modelling trains in minutes on a laptop. Datasets larger than memory train out of core, straight from disk.