twelve years after

Many seemingly different problems in artificial intelligence, symbolic processing, and machine learning can be viewed as requiring discovery of a computer program that produces some desired output for particular inputs. When viewed in this way, the process of solving these problems becomes equivalent to searching a space of possible computer programs for a most fit individual computer program. The new genetic programming paradigm described herein provides a way to search for this most fit individual computer program. In this new genetic programming paradigm, populations of computer programs are genetically bred using the Darwinian principle of survival of the fittest and using a genetic crossover (recombination) operator appropriate for genetically mating computer programs. In this paper, the process of formulating and solving problems using this new paradigm is illustrated using examples from various areas.

ions are made over the structural information (relations)

Autonomous systems require the ability to plan effective courses of action under potentially uncertain or unpredictable contingencies. Planning requires knowledge of the environment that is accurate enough to allow reasoning about actions. If the environment is too complex or very dynamic, goal-driven learning with reactive feedback becomes a necessity. This chapter addresses the issue of learning by experimentation as an integral component of PRODIGY. PRODIGY is a flexible planning system that encodes its domain knowledge as declarative operators, and applies the operator refinement method to acquire additional preconditions or postconditions when observed consequences diverge from internal expectations. When multiple explanations for the observed divergence are consistent with the existing domain knowledge, experiments to discriminate among these explanations are generated. The experimentation process isolates the deficient operator and inserts the discriminant condition or unforeseen side-effect to avoid similar impasses in future planning. Thus, experimentation is demand-driven and exploits both the internal state of the planner and any external feedback received. A detailed example of integrated experiment formulation in presented as the basis for a systematic approach to extending an incomplete domain theory or correcting a potentially inaccurate one.

After identifying essential student modeling issues and machine learning approaches, this paper examines how machine learning techniques have been used to automate the construction of student models as well as the background knowledge necessary for student modeling. In the process, the paper sheds light on the difficulty, suitability and potential of using machine learning for student modeling processes, and, to a lesser extent, the potential of using student modeling techniques in machine learning.

In this paper we examine the role of a bias toward simplicity in directing the process of representation change. We focus on the task of inducing context-free grammars from sample sentences, and we present a rational reconstruction of Wolff's SNPR -- the Grids system -- that incorporates the simplicity bias. The basic induction method alternates between merging existing nonterminal symbols and creating new symbols, using hill-climbing search to move from complex to simpler grammars. In the process, the algorithm creates word classes, phrases, and recursive rewrite rules that move beyond the training sentences, using the simplicity metric to direct search and negative training cases to eliminate overly general grammars. Experiments reveal that this approach can induce accurate grammars and that it scales reasonably to more difficult domains. Moreover, comparative studies show that both the simplicity bias and negative instances play an important role in constraining search, and that sim...

This paper proposes a new connectionist approach to numeric law discovery; i.e., neural networks (law-candidates) are trained by using a newly invented second-order learning algorithm based on a quasi-Newton method, called BPQ, and the MDL criterion selects the most suitable from law-candidates. The main advantage of our method over previous work of symbolic or connectionist approach is that it can efficiently discover numeric laws whose power values are not restricted to integers. Experiments showed that the proposed method works well in discovering such laws even from data containing irrelevant variables or a small amount of noise. 1.

A framework for knowledge-based scientific discovery in geological databases has been developed. The discovery process consists of two main steps: context definition and equation derivation. Context de nition properly defines and formulates homogeneous regions, each of which is likely to produce a unique and meaningful analytic formula for the goal variable. Clustering techniques and a suite of visualization and interpretation routines make up a tool box that assists the context definition task. Within each context, multi-variable regression analysis is conducted to derive analytic equations between the goal variable and a set of relevant independent variables, starting with one or more of the initial base models. Domain knowledge, plus a heuristic search technique called component plus residual plots dynamically guide the equation refinement process. The methodology has been applied to derive porosity equations for data collected from oil elds in the Alaska Basin. Preliminary results demonstrate the effectiveness of this methodology.

1 An understanding of learning -- the process by which a learner acquires and refines a broad range of knowledge and skills -- is central to the enterprise of building truly adaptive, flexible, robust, and creative intelligent systems. Significant theoretical and empirical contributions to the characterization of learning in computational terms have emerged from research in a number of disparate research paradigms. The limitations of individual paradigms and of particular classes of techniques within each paradigm are beginning to be recognized. Converging lines of evidence from multiple sources, both theoretical as well as empirical, suggest that artificial intelligence systems, in order to be able to deal with complex tasks such as recognizing and describing 3-dimensional objects, or communicating in natural language, must be able to effectively utilize a range of learning algorithms operating with an adequate repertoire of representational structures. This paper draws on a broad ran...

An important problem in economics and other areas of science is finding the mathematical relationship between the empirically observed variables measuring a system. In many conventional modeling techniques, one necessarily begins by selecting the size and shape of the mathematical model. Because the the vast majority of available mathematical tools only handle linear models, this choice is often simply a linear model. After making this choice, one usually then tries to find the values of certain coefficients and constants required by the particular model so as to achieve the best fit between the observed data and the model. But, in many cases, the most important issue is the size and shape of the mathematical model itself. That is, one really wants first to find the functional form of the model that best fits observed empirical data, and, only then, go on to find any constants and coefficients that happen to be needed. Some techniques exist for doing this. We suggest that finding the functional form of the model can productively be viewed as being equivalent to searching a space of possible computer programs for the particular individual computer program which produces the desired output for given inputs. That is, one is searching for the computer program