In this paper, we revisit the problem of inducing a process model from time-series data.

Research on the computational discovery of numeric equations has focused on constructing laws from scratch, whereas work on theory revision has emphasized qualitative knowledge. In this paper, we describe an approach to improving scienti c models that are cast as sets of equations. We review one such model for aspects of the Earth ecosystem, then recount its application to revising parameter values, intrinsic properties, and functional forms, in each case achieving reduction in error on Earth science data while retaining the communicability of the original model. After this, we consider earlier work on computational scienti c discovery and theory revision, then close with suggestions for future research on this topic.

In this paper we distinguish between two computational paradigms for knowledge discovery that share the notion of heuristic search, but dier in the importance they place on using scientific formalisms to state discovered knowledge. We also report progress on computational methods for discovering such communicable knowledge in two domains, one involving the regulation of photosynthesis in phytoplankton and the other involving carbon production by vegetation in the Earth ecosystem. In each case, we describe a representation for models, methods for using data to revise existing models, and some initial results. In closing, we discuss related work on the computational discovery of communicable scientific knowledge and outline directions for future research.

Abstract not found

Abstract. Using a new method published by the first author, this chapter shows how knowledge in the form of a ranked data relationship and an induced rule can be directly extracted from each training case for a Multilayer Perceptron (MLP) network with binary inputs. The knowledge extracted from all training cases can be used to validate the MLP network and the ranked data relationship for any input case provides direct user explanations. The method is demonstrated for example training cases from a real-world MLP that classifies low back pain patients into three diagnostic classes. In using the method to validate the network a number of test cases apparently mis-classified by the network were found to have most likely been incorrectly classified by the clinicians. The method uses a direct approach which does not depend on combinatorial search and is thus applicable to realworld networks with large numbers of input features, as demonstrated in this current study. 1

Ecosystem models are used to interpret and predict the interactions of different species among themselves and with their environment. In this paper, we address the task of...

Abstract. In order to discover relevant weights of neural networks, this paper proposes a novel method to learn a distinct squared penalty factor for each weight as a minimization problem over the cross-validation error. Experiments showed that the proposed method works well in discovering a polynomial-type law even from data containing irrelevant variables and a small amount of noise. 1

Ecosystem models are used to interpret and predict the interactions of species and their environment. In this paper, we address the task of inducing ecosystem models from background knowledge and timeseries data, and we review IPM, an algorithm that addresses this problem.