The state of Inductive Process Modeling (eg, SC-IPM)
has advanced considerably in the past decade,
the basic motivation and approach remain largely unchanged
from those found in the original paper:
1. Introduction and Motivation
Many scientific and engineering domains
involve continuous variables that change over time.
The increasing availability of such systems
presents both an opportunity and a challenge for machine learning.
Successful applications of induction methods hold obvious benefits,
and there exist large literatures on computational methods
for regression and time-series prediction.
But however accurate the predictive models
these techniques induce from data,
they usually make little contact with the formalisms and concepts
used by scientists and engineers.
And as Pazzani et al. (2001) have shown,
experts in some domains will reject a learning system's output,
even when very accurate, unless it makes contact with their prior knowledge.
Research on discovering numeric laws (e.g., Langley, 1981; Washio et al., 2000)
addresses this concern, in that many scientists find equations familiar.
However, although the resulting knowledge generalizes beyond the training data,
it is usually descriptive in that it directly relates observable variables.
In contrast, models in science and engineering often provide an explanation
which includes variables, objects, or mechanisms that are unobserved,
but that help predict the behavior of observed variables.
Moreover, explanations often make use of general concepts or relations
that occur in general models.
One example is Newton's theory of gravitation,
which moved beyond Kepler's descriptive laws
to an explanation of planetary trajectories
in terms of straight line motion and attractive force.
We claim that explanations in science and engineering
are often stated in terms of generic processes from some domain.
We will focus here on a particular class of processes
that describe one or more causal relations
between input variables and output variables.
A process states these relations in terms of differential equations
(for a process that involves change over time)
or static equations (for one that involves instantaneous effects).
A process may also include conditions,
stated as threshold tests on its input variables,
that describe when it is active.
A process model consists of a set of processes
that link observable input variables with observable output variables,
possibly through unobserved theoretical terms.
Inducing Process Models from Continuous Data
Langley, P., Sanchez, J., Todorovski, L., & Dzeroski, S. (2002)
Proceedings of the Nineteenth International Conference
on Machine Learning (pp. 347-354).
Sydney, Australia: Morgan Kaufmann.
For more information, see the IPM
, and Timeline
This wiki page is maintained by Rich Morin
an independent consultant specializing in software design, development, and documentation.
Please feel free to email
comments, inquiries, suggestions, etc!