Training a multilayer perceptron (MLP) with algorithms employing global search strategies has been an important research direction in the field of neural networks. Despite a number of significant results, an important matter concerning the bounds of the search region---typically defined as a box---where a global optimization method has to search for a potential global minimizer seems to be unresolved. The approach presented in this paper builds on interval analysis and attempts to define guaranteed bounds in the search space prior to applying a global search algorithm for training an MLP. These bounds depend on the machine precision and the term guaranteed denotes that the region defined surely encloses weight sets that are global minimizers of the neural network's error function. Although the solution set to the bounding problem for an MLP is in general non-convex, the paper presents the theoretical results that help deriving a box which is a convex set. This box is an outer approximation of the algebraic solutions to the interval equations resulting from the function implemented by the network nodes. An experimental study using well known benchmarks is presented in accordance with the theoretical results.

Reliable estimation of a neural network's domain of validity is important for a number of reasons such as assessing its ability to cope with a given problem, evaluating the consistency of its generalization etc. In this paper we introduce a new approach to estimate the domain of validity of a neural network based on Set Inversion Via Interval Analysis (SIVIA), the methodology established by Jaulin and Walter [1]. This approach was originally introduced in order to solve nonlinear parameter estimation problems in a bounded error context and proved to be effective in tackling several types of problems dealing with nonlinear systems analysis. The dependence of a neural network output on the pattern data is a nonlinear function and hence derivation of the impact of the input data to the neural network function can be addressed as a nonlinear parameter estimation problem that can be tackled by SIVIA. We present concrete application examples and show how the proposed method allows to delimit the domain of validity of a trained neural network. We discuss advantages, pitfalls and potential improvements offered to neural networks.