Learning Declarative Bias
In this paper, we introduce an inductive logic programming approach to learning declarative bias. The target learning task is inductive process modeling, which we briefly review. Next we discuss our approach to bias induction while emphasizing predicates
- PDF / 384,288 Bytes
- 15 Pages / 430 x 660 pts Page_size
- 34 Downloads / 228 Views
Abstract. In this paper, we introduce an inductive logic programming approach to learning declarative bias. The target learning task is inductive process modeling, which we briefly review. Next we discuss our approach to bias induction while emphasizing predicates that characterize the knowledge and models associated with the HIPM system. We then evaluate how the learned bias affects the space of model structures that HIPM considers and how well it generalizes to other search problems in the same domain. Results indicate that the bias reduces the size of the search space without removing the most accurate structures. In addition, our approach reconstructs known constraints in population dynamics. We conclude the paper by discussing a generalization of the technique to learning bias for inductive logic programming and by noting directions for future work. Keywords: inductive process modeling, meta-learning, transfer learning.
1
Introduction
Research on inductive process modeling [1] emphasizes programs that build models of dynamic systems. As the name suggests, the models are sets of processes that relate groups of entities. For example, neighboring wolf and rabbit populations interact through a predation process, which may take one of many forms. As input, these programs take observations, which record system behavior over time, background knowledge, which consists of scientifically meaningful generic processes, and entities whose behavior should be explained. The output is a model that comprises processes instantiated with the available entities. A naive solution to the task would exhaustively search the space of models defined by the instantiated processes, but this approach produces several nonsensical models and the search space grows exponentially in the number of instantiations. To make inductive process modeling manageable in nontrivial domains, one must introduce bias. Recently, researchers developed the notion of a process hierarchy to define the space of plausible model structures [2]. This solution defines which processes H. Blockeel et al. (Eds.): ILP 2007, LNAI 4894, pp. 63–77, 2008. c Springer-Verlag Berlin Heidelberg 2008
64
W. Bridewell and L. Todorovski
must always appear in a model, which ones depend on the presence of others, and which ones mutually exclude each other. Although one can use the hierarchy to substantially reduce the size of the search space, specifying relationships that both constrain the space and have validity in the modeled domain is difficult. Importantly, the introduction of this bias replaces the task of manually building a model with that of manually defining the space of plausible model structures. Ideally we would like to automatically discover this knowledge. Ample literature exists on bias selection [3], which emphasizes search through the space of learning parameters, and constructive induction [4], which increases the size of the search space. In contrast, we wish to learn constraints that will reshape the search space and ensure that the program considers only plausib
Data Loading...
