Laurence C. Dixon

Learn More
RAM-based neural networks are designed to be readily and eeciently implemented in hardware. The desire to retain this property innuences the design of learning algorithms. Reverse diieren-tiation enables high-order directional derivatives to be obtained eeciently on an arbitrary architecture. To compare the performance of diier-ent learning algorithms, four(More)
The HyperNet (Hypercube based neural Network) architecture uses higher-order sigma-pi units which are realisable with conventional memory technology. Real-valued node input and output vectors promote automatic generalisation and allow the architecture to interface to analogue data. Hardware implementation of real-valued inputs as parallel stochastic bit(More)
Current environmental monitoring systems assume particles to be spherical, and do not attempt to classify them. A laser-based system developed at the University of Hertfordshire aims at classifying airborne particles through the generation of two-dimensional scattering profiles. The pedormances of template matching, and two types of neural network (HyperNet(More)
Current environmental monitoring systems assume particles to be spherical, and do not attempt to classify them. A laser-based system developed at the University of Hertfordshire aims at classifying airborne particles through the generation of two-dimensional scattering prooles. The performances of template matching, and two types of neural network (HyperNet(More)
  • 1