Filters
Availability
Publication Year
Resource Type
Faculty
School/Institute
Publication Search Results
Sort
Results per page
1 - 2 of 2
-
(1998)Conference PaperThis paper describes experiments on on the robustness of tensor product networks using distributed representations, for recall tasks. The results of the experiments indicate, among other things, that the degree of robustness increases with the number of binding units and decreases with the fraction of the space of possible facts that have been taught to the network. Mean recall scores decrease linearly with the proportion of binding units inactivated, and recall score variance depends linearly on number of binding units and on number of facts taught to the network.
-
(1995)Conference PaperThis paper concerns a class of recurrent neural networks related to Elman networks (simple recurrent networks) and Jordan networks and a class of feedforward networks architecturally similar to Waibel’s TDNNs. The recurrent nets used herein, unlike standard Elman/Jordan networks, may have more than one state vector. It is known that such multi-state Elman networks have better learning performance on certain tasks than standard Elman networks of similar weight complexity. The task used involves learning the graphotactic structure of a sample of about 400 English words. Learning performance was tested using regimes in which the state vectors are, or are not, zeroed between words: the former results in larger minimum total error, but without the large oscillations in total error observed when the state vectors are not periodically zeroed. Learning performance comparisons of the three classes of network favour the feedforward nets.