Heuristic Induction of Rate-Based Process Models

Abstract

This paper presents a novel approach to inductive process modeling, the task of constructing a quantitative account of dynamical behavior from time-series data and background knowledge. We review earlier work on this topic, noting its reliance on methods that evaluate entire model structures and use repeated simulation to estimate parameters, which together make severe computational demands. In response, we present an alternative method for process model induction that assumes each process has a rate, that this rate is determined by an algebraic expression, and that changes due to a process are directly proportionalto its rate. We describe RPM, an implemented system that incorporates these ideas, and we report analyses and experiments that suggest it scales well to complex domains and data sets. In closing, we discuss related research and outline ways to extend the framework.

The Computational Gauntlet of Human-Like Learning

“…The question is whether human-like approaches that directly acquire modular structures in an incremental, piecemeal manner will fare better on such problems. In addition, many tasks that require parameter estimation, such as equation discovery in scientific domains, also involve creating symbolic structures (e.g., Langley and Arvay 2015), so the two paradigms are not mutually exclusive.…”

Flexible Model Induction through Heuristic Process Discovery

“…The framework has been applied to aquatic ecosystems (Asgharbeygi et al 2006), hydrology (Bridewell et al 2008), and biochemistry (Langley et al 2006). Recent work (Langley and Arvay 2015) has introduced a more constrained notation for process models, similar to that shown in Table 1, that keeps rate expressions distinct from the derivatives that are proportional to them. Briefly, their RPM system calculates, for each time step, the derivative for each variable and the rates for candidate processes; it then uses multiple linear regression to estimate the coefficients for hypothesized sets of processes that map onto model equations.…”

“…Finally, many of our runs use synthetic noise-free data from known target models. This issue is less problematic, as Langley and Arvay (2015) have shown that simple smoothing of trajectories let systems like SPM handle up to ten percent noise.…”