No coin nor oath required. For personal study only.
Reading aloud is among the problems that cannot be easily solved by conventional computing methods. NETtalk (Sejnowski and Rosenberg, 1986) demonstrated that it is possible for a parallel network of computing units to be trained to form intemal representations of the regularities in a training set. The NETtalk experiment opens the door to a host of questions like what kind of network architecture is really suited to problems of this nature, what learning strategies could be used. In particular is it possible to devise a system based on distributed representations that will be able to not only form abstractions of regularities in the training set but also translate these to other test data to show equally good generalization.
Sparse distributed memory or SDM is a distributed model of memory proposed by Kanerva (1984). It is capable of handling enormously large address spaces and shows the properties of associative recall in the presence of noise. The actualization of the memory is attained through an actualization of a small subset of the address space. This subset is a random sample of the address space.
๐ SIMILAR VOLUMES