About

History

In 2004, research within the (former) COMS research group, led by professor Tom Dhaene, was focused on developing efficient, adaptive and accurate algorithms for polynomial and rational modeling of linear time-invariant (LTI) systems. This work resulted in a set of Matlab scripts that were used as a testing ground for new ideas and concepts. Research progressed, and with time these scripts were re-worked and refactored into one coherent Matlab toolbox, tentatively named the Multivariate MetaModeling (M3) Toolbox. The first public release of the toolbox (v2.0) occurred in November 2006. In October 2007, the development of the M3 Toolbox was discontinued.

In April 2008, the first public release of the Surrogate Modeling (SUMO) Toolbox (v5.0) occurred.

For a list of changes since then refer to the Changelog.

Intended use

Global Surrogate Models

The SUMO Toolbox was designed to solve the following problem:

Automatically generate a highly accurate surrogate model (= a regression model) for a computational expensive simulation code requiring as little data points and as little user-interaction as possible.

In addition the toolbox provides powerful, adaptive algorithms and a whole suite of model types for

• data fitting problems (regression, function approximation)
• response surface modeling (RSM)
• interpolation
• model selection
• Design Of Experiments (DoE)
• model parameter optimization (hyperparameter selection)
• adaptive sample selection (also known as sequential design or active learning)

For application scientists or engineers the SUMO Toolbox provides a flexible, pluggable platform to which the response surface modeling task can be delegated. For researchers in surrogate modeling it provides a common framework to implement, test and benchmark new modeling and sampling algorithms.

See the Wikipedia surrogate model page to find out more.

Surrogate Driven Optimization

While the main focus of the SUMO Toolbox is to create accurate global surrogate models, it can be used for other goals too.

For instance, the toolbox can be used to create consecutive local surrogate models for optimization purposes. The information obtained from the local surrogate models is used to guide the adaptive sampling process to the global optimum.

A good sample strategy for surrogate driven optimization seeks a balance between local search and global search, or refining the surrogate model and finding the optimum. Such a sample strategy is implemented (akin to (Super)EGO), see the different sample selectors for more information.

Application range

The SUMO Toolbox has already been applied successfully to a wide range of problems from domains as diverse as aerodynamics, geology, metallurgy, electro-magnetics (EM), electronics, engineering and economics.

Throughout the different problems, the input dimension has ranged from 1 to 130 and the output dimension from 1 to 70 (including both complex and real valued outputs). The number of data points has ranged from as little as 15 to as many as 100000.

Design goals

During research into multivariate surrogate modeling techniques and algorithms it became clear that there was room for an adaptive tool that integrated different surrogate modeling approaches and did not tie the user down to one particular set of problems or techniques.

More concretely, we were unable to find evidence of any projects that integrated:

1. Building standalone global surrogate models (= replacement metamodels)
2. Support for different model types, different model parameter optimization algorithms, different model selection criteria, ... (adaptive modeling)
3. Sequential design (selecting data points iteratively and pro-actively)
4. Distributed computing (integration with cluster and grid middleware to transparently run simulations in parallel)
5. Usable implementation in software

This gave rise to a number of design goals that served as the guidelines for the design of the SUMO Toolbox.

These goals are:

1. Development of a fully automated, adaptive surrogate model construction algorithm. Given a simulation model, the software should produce a replacement metamodel with as little user interaction as possible ("one button approach").
2. There is no such thing as a "one-size-fits-all", different problems need to be modeled differently and require different levels of process knowledge. Therefore the software should be modular and extensible but not be too cumbersome to use or configure (sensible defaults).
3. The toolbox should minimize the required prior knowledge of the system to be modeled.
4. The algorithm should minimize the number of required data samples in order to come to an acceptable surrogate model.
5. The algorithm should terminate only when the predefined accuracy (set by the user) has been reached or the maximum number of iterations/samples has been exceeded.

Features

The main features of the toolbox are listed below. For an overview of recent changes see the Whats new page. A detailed list of changes can be found in the Changelog.

Implementation Language	Matlab, Java, and where applicable C, C++
Design patterns	Fully object oriented, with the focus on clean design and encapsulation.
Minimum Requirements	See the system requirements page
Supported data sources*	Local executable/script, simulation engine, Java class, Matlab script, dataset (txt file) (see Data format)
Supported data types	Supports multi-dimensional inputs and outputs. Outputs can be any combination of real/complex.
Configuration	Extensively configurable through one main XML configuration file.
Flexibility	Virtually every component of the modeling process can be configured, replaced or extended by a user specific, custom implementation
Predefined accuracy	The toolbox will run until the user required accuracy has been reached (on the selected measures), the maximum number of samples has been exceeded or a timeout has occurred
Model Types*	Out of the box support for: Polynomial/Rational functions Feedforward Neural Networks, 3 implementations (the last two since v5.0) One based on the Matlab Neural Network toolbox One based on the Fast Artificial Neural Network Library (FANN) One based on the NNSYSID Toolbox Radial Basis Function (RBF) Models RBF Neural Networks Kriging Models, 2 implementations: One custom implementation One based on the DACE toolbox Smoothing spline models Support Vector Machines (SVM) Least Squares SVM (based on LS-SVMlab) epsilon-SVM (based on LIBSVM or SVMlight) nu-SVM (based on LIBSVM)
Model parameter optimization algorithms*	Pattern Search, Simulated Annealing, Genetic Algorithm, BGFS, DIRECT, Particle Swarm Optimization (PSO), ...
Sample selection algorithms (=sequential design, active learning)*	Random, error-based, density-based, gradient-based
Experimental design*	Latin Hypercube Sampling, Central Composite, Box-Behnken, random, dataset based, full factorial, adaptive (by doing a preliminary 1D screening in each dimension)
Model selection measures*	Validation set, cross-validation, leave-one-out, comparison on a grid, AIC
Sample Evaluation*	On the local machine (taking advantage of multi-core CPUs) or in parallel on a cluster/grid
Supported distributed middlewares*	Sun Grid Engine, LCG Grid middleware (both accessed through a SSH accessible frontnode)
Logging	Extensive logging to enable close monitoring of the modeling process. Logging granularity is fully configurable and log streams can be easily redirected (to file, console, a remote machine, ...).
Profiling*	Extensive profiling framework for easy gathering (and plotting) of modeling metrics (average sample evaluation time, hyperparameter optimization trace, ...)
Easy tracking of modeling progress	Automatic storing of best models and their plots. Ability to automatically generate a movie of the sequence of plots.
Available test problems*	Out of the box support for various built-in functions (Ackley, Camel Back, Goldstein-Price, ...) and datasets (Abalone, Boston Housing, FishLength, ...) from various application domains. Including a number of datasets (and some simulation code) from electronics. In total over 50 examples are available.

* Custom implementations can easily be added

Screenshots

A number of screenshots to give a feel of the SUMO Toolbox.

• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 

Note these screenshots do not necessarily reflect the latest toolbox version.

Movies

A number of movies that illustrate how the modeling process progresses as more samples come in.

• Modeling the Step-Discontinuity (= electro-magnetic problem)
  • rational model
• Modeling the Ackley function (= mathematical function)
  • RBF model
  • LS-SVM model
• Particle Swarm Optimization in the parameter space of Kriging (theta)
  • PSO movie, this movie shows how the correlation parameters are optimized as the SUMO toolbox searches for better models. Note that the data distribution is not constant, but is continually updated.
• Modeling David's face (Data courtesy of the Digital Michelangelo Project)
  • RBF model

A number of movies created with the Model Visualization GUI.

• Visualizing a 3D model from video compression data
  • RBF model
• Modeling an exponential tapered TML (described in Microwave Engineering 2nd Edition, D.M. Pozar) using an Artificial Neural Network
  • ANN model
• Visualizing 3 (out of 5) input dimensions with a high framerate and lighting, no samples shown.
  • ANN model
• A movie of a 3D model with sample points, but without lighting.
  • rational model
• Visualizing a 3D model on a contour plot.
  • rational model
• A movie of a 2D model on a 1D plot.
  • rational model

Note these movies do not necessarily reflect the latest toolbox version. Improvements and/or interface adjustments may have been made since then.

Documentation

Presentation

• Poster: SUMO poster
• Presentation: SUMO slides

Newsletter

To stay up to date with the latest news and releases, we also recommend subscribing to the SUMO newsletter.

Traffic will be kept to a minimum and you can unsubscribe at any time.

References

See Citing the toolbox.