from data to knowledge

Brief description

The SUMO Toolbox is a Matlab toolbox that automatically builds accurate surrogate models (also known as metamodels or response surface models) of a given data source (e.g., simulation code, data set, script, ...) within the accuracy and time constraints set by the user. The toolbox minimizes the number of data points (which it selects automatically) since they are usually expensive.

Key features

  1. Efficiency
    • Fast design space exploration: compact scalable regression models for design automation, parametric studies, design space exploration, optimization, yield improvement, visualization, prototyping, and sensitivity analysis

    • Gain insight: knowledge discovery in sparse data sets, and knowledge extraction from large data sets

  2. Accuracy
    • Best-in-class modeling techniques: highly accurate and efficient proprietary and state-of-the-art surrogate modeling algorithms

  3. Ease-of-use
    • Expert know-how at your fingertips: sensible default settings, based on expert knowledge from various disciplines (e.g., machine learning, approximation theory, numerical analysis, statistics, optimization, ...), and also many expert options available

    • Powerful logging and profiling tools: intermediate models (and plots) stored for further reference, extensive logging of what is going on, profiling framework to track modeling progress

  4. Automation
    • Active learning: automatic selection of data points (also known as adaptive sample selectionsequential design, or optimal experimental design)

    • Model selection: automatic selection of model type (e.g., ANN, SVM, rational model, ...) and model complexity (e.g., number of neurons and hidden layers, kernel function, order, ...)

  5. Flexibility
    • Pluggable and extensible framework: easy integration of custom implementations (e.g., sampling strategies, model types, model selection criteria, hyperparameter optimization algorithms,...)

    • Flexible experimental environment: easy to setup and run different modeling experiments, easy to benchmark different techniques

    • Multi-platform: available for Windows, Mac OSX, and Unix/Linux platforms

  6. Speed
    • Shorten time to market: lower cost and shorten process cycle time

    • Distributed computing: integration with cluster and grid middleware to transparently run simulations in parallel