Prediction- and simulation-error based perceptron training: Solution space analysis and a novel combined training scheme

Previous papers have noted the difficulty in obtaining neural models which are stable under simulation when trained using prediction-error-based methods. Here the differences between series-parallel and parallel identification structures for training neural models are investigated. The effect of the error surface shape on training convergence and simulation performance is analysed using a standard algorithm operating in both training modes. A combined series-parallel/parallel training scheme is proposed, aiming to provide a more effective means of obtaining accurate neural simulation models. Simulation examples show the combined scheme is advantageous in circumstances where the solution space is known or suspected to be complex. (c) 2006 Elsevier B.V. All rights reserved.

Improving neural network training solutions using regularisation

This paper describes the application of regularisation to the training of feedforward neural networks, as a means of improving the quality of solutions obtained. The basic principles of regularisation theory are outlined for both linear and nonlinear training and then extended to cover a new hybrid training algorithm for feedforward neural networks recently proposed by the authors. The concept of functional regularisation is also introduced and discussed in relation to MLP and RBF networks. The tendency for the hybrid training algorithm and many linear optimisation strategies to generate large magnitude weight solutions when applied to ill-conditioned neural paradigms is illustrated graphically and reasoned analytically. While such weight solutions do not generally result in poor fits, it is argued that they could be subject to numerical instability and are therefore undesirable. Using an illustrative example it is shown that, as well as being beneficial from a generalisation perspective, regularisation also provides a means for controlling the magnitude of solutions. (C) 2001 Elsevier Science B.V. All rights reserved.

Training of Reaching in Stroke Survivors With Severe and Chronic Upper Limb Paresis Using a Novel Nonrobotic Device: A Randomized Clinical Trial

Background and Purpose—Severe upper limb paresis is a major contributor to disability after stroke. This study investigated the efficacy of a new nonrobotic training device, the Sensorimotor Active Rehabilitation Training (SMART) Arm, that was used with or without electromyography-triggered electrical stimulation of triceps brachii to augment elbow extension, permitting stroke survivors with severe paresis to practice a constrained reaching task.

Methods—A single-blind, randomized clinical trial was conducted with 42 stroke survivors with severe and chronic paresis. Thirty-three participants completed the study, of whom 10 received training using the SMART Arm with electromyography-triggered electrical stimulation, 13 received training using the SMART Arm alone, and 10 received no intervention (control). Training consisted of 12 1-hour sessions over 4 weeks. The primary outcome measure was “upper arm function,” item 6 of the Motor Assessment Scale. Secondary outcome measures included impairment measures; triceps muscle strength, reaching force, modified Ashworth scale; and activity measures: reaching distance and Motor Assessment Scale. Assessments were administered before (0 weeks) and after training (4 weeks) and at 2 months follow-up (12 weeks).

Results—Both SMART Arm groups demonstrated significant improvements in all impairment and activity measures after training and at follow-up. There was no significant difference between these 2 groups. There was no change in the control group.

Conclusions—Our findings indicate that training of reaching using the SMART Arm can reduce impairment and improve activity in stroke survivors with severe and chronic upper limb paresis, highlighting the benefits of intensive task-oriented practice, even in the context of severe paresis.