242 S. Stijven et al.
the model development is a huge advantage of multi-objective symbolic regression
and provides the foundations for the use cases which are discussed in this chapter.
We open the discussion with a brief review of the role of symbolic regression in
process optimization. Although such is one of the earliest real-world applications of
GP, success in this space fueled a sizable investment in developing new algorithms,
analysis infrastructure and workflow which, subsequently, enabled application in
other domains.
The maturation of the capabilities has enabled a migration away from the world
of corporate R&D and into the business mainstream. The business forecasting
application discussed in the second section features the hybridization and integration
of symbolic regression with other technologies—in this case, ARIMAX model-
ing—which is facilitated by the white-box nature of the developed models.
The final sections are devoted to the development of a metamodel (or a surrogate
model) to summarize the propagation of infectious disease for purposes of social
policy design. As such it makes the results of highly sophisticated simulation models
accessible to policy makers who, via an interactive interface, explore the impact
and implications of different assumptions and scenarios. The immediacy of the
interaction enables the an awareness and integration of insight which would not
be possible if review of myriad charts, tables and text were required.
In summary, over the course of a quarter-century, symbolic regression has moved
from the realm of the research lab into manufacturing to business operations and
policymaking. This indicates a level of maturity such that it is ready to take its
proper place among the pantheon of data analysis tools
2 Modern Process Optimization
2.1 Historical Foundations
One of the first applications of genetic programming in Dow Chemical was in
continuous process modeling (Smits and Kotanchek 2004 ; Kordon and Smits 2001 ).
Over the subsequent 25 years, orders-of-magnitude improvements in both the
algorithmic efficiency of symbolic regression as well as corresponding orders-of-
magnitude enhancements in compute capabilities has greatly expanded the impact
and efficacy of symbolic regression in this realm. In parallel with the model
generation capability improvements, new tools and techniques for both designing
the modeling strategy and extracting insight from the new plethora of developed
models have greatly enhanced the impact. In many respects, the overarching analysis
workflow improvement has been as meaningful as the improvement in the model
development foundations.