Frank J. Smieja

Learn More
There are currently several types of constructive, or growth, algorithms available for training a feed-forward neural network. This paper describes and explains the main ones, using a fundamental approach to the multi-layer perceptron problem-solving mechanisms. The claimed convergence properties of the algorithms are veriied using just two mapping theorems(More)
Learning from examples has a number of distinct algebraic forms, depending on what is to be learned from which available information. One of these forms is x G ! y, where the input{output tuple (x; y) is the available information , and G represents the process determining the mapping from x to y. Various models , y = f(x), of G can be constructed using the(More)
Adaptive models of systems seek to emulate the processes giving rise to the data observed in the system. The process is often termed learning from examples, or data-driven information processing. An important issue regarding such modeling is the active selection of data by the modeling process, or exploration. If exploration depends on the current state of(More)
Learning from examples has a number of distinct algebraic forms, depending on what is to be learned from which available information. One of these forms is x G ! y, where the input{output tuple (x; y) is the available information, and G represents the process determining the mapping from x to y. Various models, y = f(x), of G can be constructed using the(More)
Inversion of the kinematics of ma-nipulators is one of the central problems in the eld of robot arm control. The iterative use of inverse diierential kinematics is a popular method of solving this task. Normally the solution of the problem requires a complex mathematical apparatus. It involves methods for solving equation systems as well as algorithms for(More)
An extended feed-forward algorithm for recurrent connectionist networks is presented. This algorithm, which works locally in time, is derived both for discrete-in-time networks and for continuous networks. Several standard gradient descent algorithms for connectionist networks (e.g. 48], 30], 28] 15], 34]), especially the backpropagation algorithm 36], are(More)
Many of the current artiicial neural network systems have serious limitations, concerning accessibility , exibility, scaling and reliability. In order to go some way to removing these we suggest a reeective neural network architecture. In such an architecture, the modular structure is the most important element. The building-block elements are called(More)