Specialists and non-specialists in artificial neural networks (ANN) must closely interact in many applications, including structural sensing. The non-specialists must be aware of ANN-specific terminology, capabilities, and connecting concepts for effective collaboration. An instructional approach for ANNs is described that progresses from practical concepts to guided MatLab-based experimentation. Back propagation-trained multilayer perceptron neural networks are presented with an emphasis on parallel processing and training characteristics. The one-week instructional module has a lecture to convey terminology and structure, detailed examples to illustrate the training process, and guided application-based exercises. The MatLab neural-networks toolbox provides a transparent learning environment in which the students focus on network design and training concepts rather than the tool itself. Learning effectiveness was evaluated in an applications-oriented sensors curriculum. Instructional resources including realistic problems are webaccessible. These resources may be adjusted for different degrees of challenge and for simpler or more realistic problem solving.
National Science Foundation (U.S.)
- Artificial Neural Networks (ANN),
- Sensors Curriculum
Available at: http://works.bepress.com/rohit-dua/1/