Speaker:

Pat Langley
Distinguished Scientist, CMU SV; Professor of Computer Science, University of Auckland

Date/Time:

November 13, 1:30 pm

Location:

Webcast:

CMUSV, Rm 118 [directions]

[login as guest]

Early research in e-science emphasized representing and simulating models that reflected scientists' knowledge, but these models often made little contact with data. Recent work in e-science has utilized machine learning and data mining to uncover regularities in data, but makes few connections to scientists' knowledge. In this talk, I present an approach known as inductive process modeling that combines these two traditions. The paradigm encodes scientific models as sets of processes that incorporate differential equations, induces the models from time-series data, and uses background knowledge to guide their construction. The resulting models are interpretable, but they are also accurate, in that they match observations. I illustrate this approach in the context of ecology and environmental science, and I report extensions that increase the plausibility of induced models and efficiency at finding them. In addition, I report an interactive software environment for the construction, evaluation, and revision of such interpretable scientific models.
This talk describes joint work at Stanford University and ISLE with Kevin Arrigo, Stuart Borrett, Matt Bravo, Will Bridewell, and Ljupco Todorovski. Papers are available at http://www.isle.org/process/.