Constructive learning algorithms oﬀer an attractive approach for the incremental construction of near-minimal neural-network architectures for pattern classiﬁcation. This paper explores the feasibility of using a constructive algorithm for multilayer perceptrons (MLPs) applied to the problem of speech/non-speech classiﬁcation in hearing aids. When properly designed and trained, MLPs are able to generate an arbitrary classiﬁcation frontier with a relatively low computational complexity. The paper will focus on the design of a constructive algorithm for MLPs which attempts to converge to the minimum complexity network for the given problem. The results obtained will be compared with those cases in which the constructive algorithm is not considered.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.