Classification using hierarchical mixtures of experts

Abstract

There has recently been widespread interest in the use of multiple models for classification and regression in the statistics and neural networks communities. The Hierarchical Mixture of Experts (HME) [1] has been successful in a number of regression problems, yielding significantly faster training through the use of the Expectation Maximisation algorithm. In this paper we extend the HME to classification and results are reported for three common classification benchmark tests: Exclusive-Or, N-input Parity and Two Spirals.

From quantitative microscopy to automated image understanding

“…AdaBoost's idea of focusing a classifier on hard training examples, Mixtures-of-Experts goes one step further by training individual classifiers, also called local experts, at different data partitions and combining the results from multiple classifiers in a trainable way [45,46]. In Mixtures-of-Experts, a gating network is employed to assign local experts to different data partitions and the local experts, which can be various classifiers, take the input data and make predictions.…”

Modular neural networks for multi-service connection admission control

“…When the HME was used in time-series prediction tasks [30,37], results that were better than leading methods, including those using MLP networks [38], can be obtained. Signi®cant improvement of an existing¯ow control scheme [29] has also been achieved by the authors, using an HMEbased predictor [31].…”

Modularity in neural computing

“…By then considering only the most probable path through Table 1 Application of MNN's in Pattern Recognition Theory the tree, they pruned branches away, either temporarily or permanently in case of redundancy [138]. This improved HME showed significant speedups and more efficient use of parameters over the standard fixed HME structure in discriminating for artificial applications as well as prediction of parameterized speech over short time segments [137]. The HME architecture has also been applied to text-dependent speaker identification [21].…”