Abstract
In studying data sets for complex nonlinear relations, neural networks can be used as modeling tools. Trained fully connected networks cannot, however, reveal the relevant inputs among a large set of potential ones, so a pruning of the connections must be undertaken to reveal the underlying relations. The paper presents a general method for detecting nonlinear relations between a set of potential inputs and an output variable. The method is based on a neural network pruning algorithm, which is run repetitively to finally yield Pareto fronts of solutions with respect to the approximation error and network complexity. The occurrence of an input on these fronts is taken to reflect its relevance for describing the output variable. The method is illustrated on a simulated cell population sensitized to death-inducing ligands resulting in programmed cell death (apoptosis).
Original language | Undefined/Unknown |
---|---|
Pages (from-to) | 687–695 |
Number of pages | 9 |
Journal | Lecture Notes in Computer Science |
Volume | 6457 |
DOIs | |
Publication status | Published - 2010 |
MoE publication type | A1 Journal article-refereed |
Keywords
- apoptosis model
- data mining
- Nonlinear modeling
- pruning