A review of RAM based neural networks
J. Austin · 2002
Reviews a class of neural networks often termed 'RAM-based networks'. As this paper shows, the networks are identified by their use of 'logical' 1-in-n decoders as a pre-process to each neuron. The paper explains why the networks have also been termed weightless systems. Two sub-classes of binary neural networks are described, those which use binary weights and use only a single layer of neurons [consisting of the multi-RAM discriminator (MRD), ADAM multiprocessor and WISARD pattern recognition machine] and those which use multi-valued weights and multiple layers of neurons [comprising the probabilistic logic node (PLN), probabilistic RAM (pRAM), goal-seeking neuron (GSN), and time-integrating neuron (TIN) networks]. The paper attempts to show the evolution of the networks, as well as describing the benefits of this class of neural network for hardware implementation.