Decision making tasks that involve delayed consequences are very common yet difficult to address with supervised learning methods. If there is an accurate model of the underlying dynamical system, then these tasks can be formulated as sequential decision problems and solved by Dynamic Programming. This paper discusses reinforcement learningin terms of the sequential decision framework and shows how a learning algorithm similar to the one implemented by the Adaptive Critic Element used in the pole-balancer of Barto, Sutton, and Anderson (1983), and further developed by Sutton (1984), fits into this framework. Adaptive neural networks can play significant roles as modules for approximating the functions required for solving sequential decision problems.

We introduce a cost function for learning in feed-forward neural networks which is an explicit function of the internal representation inaddition to the weights. The learning problem can then be formulated as two simple perceptrons and a search for internal representations. Back-propagation is recovered as a limit. The frequency of successful solutions is better for this algorithm than for back-propagation when weights and hidden units are updated on the same timescale i.e. once every learning step. 1 INTRODUCTION In their review of back-propagation in layered networks, Rumelhart et al. (1986) describe the learning process in terms of finding good "internal representations" of the input patterns on the hidden units. However, the search for these representations isan indirect one, since the variables which are adjusted in its course are the connection weights, not the activations of the hidden units themselves when specific input patterns are fed into the input layer. Rather, the internal representations are represented implicitly in the connection weight values. More recently, Grossman et al. (1988 and 1989)1 suggested a way in which the search for internal representations could be made much more explicit.

We have done an empirical study of the relation of the number of parameters (weights) in a feedforward net to generalization performance.

ABSTRACT The CHAC storage scheme has been used as a basis for a software implementation of an associative .emory A major disadvantage of this CHAC-concept is that the degree of local generalization (area of interpolation) isfixed. This paper deals with an algorithm for self-organizing variable generalization for the AKS, based on ideas of T. Kohonen. 1 INTRODUCTION For several years research at the Department of Control Theory andRobotics at the Technical University of Darmstadt has been concerned with the design of a learning real-time control loop with neuron-like associative memories (LERNAS) A Self-organizing Associative Memory System for Control Applications 333 for the control of unknown, nonlinear processes (Ersue, Tolle, 1988). This control concept uses an associative memory systemAHS, based on the cerebellar cortex model CHAC by Albus (Albus, 1972), for the storage of a predictive nonlinear processmodel and an appropriate nonlinear control strategy (Fig.1). Figure 1: The learning control loop LERNAS One problem for adjusting the control loop to a process is, however, to find a suitable set of parameters for the associative memory.The parameters in question determine the degree of generalization within the memory and therefore have a direct influence on the number of training steps required tolearn the process behaviour. For a good performance of the control loop it· is desirable to have a very small generalization around a given setpoint but to have a large generalization elsewhere. Actually, the amount of collected datais small during the transition phase between two 334 Hormel setpointsbut is large during setpoint control.

We have used information-theoretic ideas to derive a class of practical andnearly optimal schemes for adapting the size of a neural network. By removing unimportant weights from a network, several improvementscan be expected: better generalization, fewer training examples required, and improved speed of learning and/or classification. The basic idea is to use second-derivative information tomake a tradeoff between network complexity and training set error. Experiments confirm the usefulness of the methods on a real-world application. 1 INTRODUCTION Most successful applications of neural network learning to real-world problems have been achieved using highly structured networks of rather large size [for example (Waibel, 1989; Le Cun et al., 1990a)]. As applications become more complex, the networks will presumably become even larger and more structured.

This article begins with an elaboration of models of design as a process. It then introduces and describes a knowledge representation schema for design called design prototypes. This schema supports the initiation and continuation of the act of designing. Design prototypes are shown to provide a suitable framework to distinguish routine, innovative, and creative design.

This article discusses building a computable design process model, which is a prerequisite for realizing intelligent computer-aided design systems. First, we introduce general design theory, from which a descriptive model of design processes is derived. In this model, the concept of metamodels plays a crucial role in describing the evolutionary nature of design. Second, we show a cognitive design process model obtained by observing design processes using a protocol analysis method. We then discuss a computable model that can explain most parts of the cognitive model and also interpret the descriptive model. In the computable model, a design process is regarded as an iterative logical process realized by abduction, deduction, and circumscription. We implemented a design simulator that can trace design processes in which design specifications and design solutions are gradually revised as the design proceeds.

I appreciated very much the Spring 1990 issue of the AI Magazine on Robotic Assembly and Task Planning. It seems to me, however, that some good work that has been carried out on this subject in Europe during recent years has not been covered very much. Also commons on the low participation levels of women in the computer industry, suggestions for the inclusion of dissertation abstracts, comments on the Feldman article in the Fall 1990 issue, and a note about the discontinuance of plastic coverings on AI Magazine.

For our purposes, the goal is to make robots that are as humanlike as possible. Now the question becomes, Could we develop these systems to the point where x/h and The question of whether a computer deep interconnections among mind x/r were used interchangeably. In this can think like a person is once again and body are the crux of the issue. Somewhat to my surprise, Two basic lines of reasoning are thing when we said that Mary or R2D2 this philosophical question used to support the notion that computers understands Proust or loves John. The more common x/r could equal x/h, we must look understanding.

The Association for the Advancement of Artificial Intelligence held its 1990 Spring Symposium Series on March 27-29 at Stanford University, Stanford, California. This article contains a short summary of seven of the nine symposia that were conducted: AI and Molecular Biology, AI in Medicine, Automated Abduction, Case Based Reasoning, and Knowledge-Based Environments for Teaching and Learning.