Using the EM Algorithm to Train Neural Networks: Misconceptions and a New Algorithm for Multiclass Classification

The expectation-maximization (EM) algorithm has been of considerable interest in recent years as the basis for various algorithms in application areas of neural networks such as pattern recognition. However, there exists some misconceptions concerning its application to neural networks. In this paper, we clarify these misconceptions and consider how the EM algorithm can be adopted to train multilayer perceptron (MLP) and mixture of experts (ME) networks in applications to multiclass classification. We identify some situations where the application of the EM algorithm to train MLP networks may be of limited value and discuss some ways of handling the difficulties. For ME networks, it is reported in the literature that networks trained by the EM algorithm using iteratively reweighted least squares (IRLS) algorithm in the inner loop of the M-step, often performed poorly in multiclass classification. However, we found that the convergence of the IRLS algorithm is stable and that the log likelihood is monotonic increasing when a learning rate smaller than one is adopted. Also, we propose the use of an expectation-conditional maximization (ECM) algorithm to train ME networks. Its performance is demonstrated to be superior to the IRLS algorithm on some simulated and real data sets.

Using "war stories" to train for adaptive performance: Is it better to learn from error or success?

The study examines whether error exposure training can enhance adaptive performance. Fifty-nine experienced fire-fighters undergoing training for incident command participated in the study. War stories were developed based on real events to illustrate successful and unsuccessful incident command decisions. Two training methodologies were compared and evaluated. One group was trained using case studies that depicted incidents containing errors of management with severe consequences in fire-fighting outcomes (error-story training) while a second group was exposed to the same set of case studies except that the case studies depicted the incidents being managed without errors and their consequences (errorless-story training). The results provide some support for the hypothesis that it is better to learn from other people's errors than from their successes. Implications for training are discussed.

Virtual training environments to improve train driver's crisis decision making

This paper reports on a current research project in which virtual reality simulators are being investigated as a means of simulating hazardous Rail work conditions in order to allow train drivers to practice decision-making under stress. When working under high stress conditions train drivers need to move beyond procedural responses into a response activated through their own problem-solving and decision-making skills. This study focuses on the use of stress inoculation training which aims to build driver’s confidence in the use of new decision-making skills by being repeatedly required to respond to hazardous driving conditions. In particular, the study makes use of a train cab driving simulator to reproduce potentially stress inducing real-world scenarios. Initial pilot research has been undertaken in which drivers have experienced the training simulation and subsequently completed surveys on the level of immersion experienced. Concurrently drivers have also participated in a velocity perception experiment designed to objectively measure the fidelity of the virtual training environment. Baseline data, against which decision-making skills post training will be measured, is being gathered via cognitive task analysis designed to identify primary decision requirements for specific rail events. While considerable efforts have been invested in improving Virtual Reality technology, little is known about how to best use this technology for training personnel to respond to workplace conditions in the Rail Industry. To enable the best use of simulators for training in the Rail context the project aims to identify those factors within virtual reality that support required learning outcomes and use this information to design training simulations that reliably and safely train staff in required workplace accident response skills.