The Evolutionary Model Builder for Extensible Regression (EMBER) was developed by PEC with support from Applied Research Associates and Synthetik Applied Technologies. EMBER is a desktop application that contains an artificial intelligence solver for conducting symbolic regression of data sets. It uses evolutionary algorithms to conduct symbolic regression of high dimensionality data sets, employing a field of artificial intelligence known as genetic programming.
The key strength of EMBER is the ability to discover the best low-order, physics-based expressions to represent the user’s data. The EMBER philosophy is that, wherever feasible, lower order functions are generally preferable to the complex functions employed by other multi-dimensional methods (radial basis functions, neural networks, etc.). Functions with fewer degrees of freedom are often more stable for reasonable extrapolations beyond the edges of the training data. If the lower order functions also capture physics relationships of the real-world system, the likelihood of extrapolation stability further increases.
PEC employs EMBER on a range of projects to build physics-based models of user data, and EMBER is available at no cost for government clients.