Seminar on Computational Learning and Adaptation


  Discovery of Partial Differential Equations: A Declarative Bias Approach

Ljupco Todorovski
Jozef Stefan Institute
Ljubljana, Slovenia
www-ai.ijs.si/~ljupco/

This talk focuses on discovery of partial differential equations, which are one of the most powerful and widely accepted analytical formalisms for modeling biological and physical systems. Establishing an acceptable partial differential equation model from example behavior of the dynamic system has two main aspects. First, an appropriate structure must be determined for the equations involved (the problem of model identification). Second, accurate values of the constant parameters must be determined (the problem of parameter estimation). We focus on the first, more challenging, aspect of modeling, drawing on the system Lagramge, which uses a declarative bias approach to equation discovery. Instead of exploring a fixed (hard-coded) space of possible equation structures, Lagramge uses a context-free grammar to define and restrict its hypothesis space. The context-free grammar is provided by the user and is based on theoretical knowledge about the domain. We demonstrate the efficacy of this approach by its rediscovery of the classic FitzHugh-Nagumo model from biology. This represents a very wide class of biological systems, making the equation discovery of interest to scientists concerned with the enterprise of obtaining a mathematical understanding of dynamic processes that occur in the life sciences.


Date: Thurs., Oct 5

Time: 4:15-5:30PM

Place: Cordura 100


Return to the seminar schedule