Ensemble MLP Classifier Design
Multi-layer perceptrons (MLP) make powerful classifiers that may provide superior performance compared with other classifiers, but are often criticized for the number of free parameters. Most commonly, parameters are set with the help of either a validati
- PDF / 284,795 Bytes
- 15 Pages / 430 x 660 pts Page_size
- 89 Downloads / 211 Views
Abstract. Multi-layer perceptrons (MLP) make powerful classifiers that may provide superior performance compared with other classifiers, but are often criticized for the number of free parameters. Most commonly, parameters are set with the help of either a validation set or crossvalidation techniques, but there is no guarantee that a pseudo-test set is representative. Further difficulties with MLPs include long training times and local minima. In this chapter, an ensemble of MLP classifiers is proposed to solve these problems. Parameter selection for optimal performance is performed using measures that correlate well with generalisation error.
1 Introduction The topic of this chapter concerns solving problems in pattern recognition using a combination of neural network classifiers. Pattern classification involves assignment of an object to one of several pre-specified categories or classes, and is a key component in many data interpretation activities. Here we focus on classifiers that learn from examples, and it is assumed that each example pattern is represented by a set of numbers, which are known as the pattern features. In the case of face recognition (Section 5), these features consist of numbers representing different aspects of facial features. In order to design a learning system it is customary to divide the example patterns into two sets, a training set to design the classifier and a test set, which is subsequently used to predict the performance when previously unseen examples are applied. A problem arises when there are many features and relatively few training examples, and the classifier can learn the training set too well, known as over-fitting so that performance on the test set degrades. Automating the classification task to achieve optimal performance has been studied in the traditional fields of pattern recognition, machine learning and neural networks as well as newer disciplines such as data fusion, data mining and knowledge discovery. Traditionally, the approach that has been used in the design of pattern classification systems is to experimentally assess the performance of several classifiers with the idea that the best one will be chosen. Ensemble classifiers, also known as Multiple Classifier Systems (MCS), were developed to address the problem of reliably designing a system with improved accuracy. Recognising that each classifier may make different and perhaps complementary errors, the idea is to pool together the results from all classifiers to find a composite system that outperforms any individual (base) classifier. In this way a single complex classifier may be replaced by a set of relatively simple classifiers. L.C. Jain et al. (Eds.): Comp. Intel. Para.: Innov. Applications, SCI 137, pp. 133–147, 2008. springerlink.com © Springer-Verlag Berlin Heidelberg 2008
134
Terry Windeatt
Even though an ensemble is less likely to over-fit, there is still the difficulty of tuning individual classifier parameters. Multi-layer perceptrons (MLP) make powerful classifiers that may provide supe
Data Loading...
