The Gamma MLP – Using Multiple Temporal Resolutions for Improved Classification

Content Provider	Semantic Scholar
Author	Lawrence, Steve Tsoi, Ah Chung
Copyright Year	1992
Abstract	We have previously introduced the Gamma MLP which is defined a s an MLP with the usual synaptic weights replaced by gamma filters and associated gain terms throughout all layers. In this paper we apply the G amma MLP to a larger scale speech phoneme recognition problem, analyze t he operation of the network, and investigate why the Gamma MLP can perform bette r than alternatives. The Gamma MLP is capable of employing multiple tempor al resolutions (the temporal resolution is defined here, as per de Vries and P rincipe, as the number of parameters of freedom (i.e. the number of tap varia bles) per unit of time in the gamma memory – this is equal to the gamma memory parameter as detailed in the paper). Multiple temporal resolutions may b e advantageous for certain problems, e.g. different resolutions may be optima l for extracting different features from the input data. For the problem in this p aper, the Gamma MLP is observed to use a large range of temporal resolutions. In comparison, TDNN networks typically use only a single temporal resoluti on. Further motivation for the Gamma MLP is related to the “curse of dimension ality” and the ability of the Gamma MLP to trade off temporal resolution for memory depth, and therefore increase memory depth without increasing the dim nsionality of the network. The IIR MLP is a more general version of the Gamma MLP – however the IIR MLP performs poorly for the problem in this pa per. Investigation suggests that the error surface of the Gamma MLP is mor e suitable for gradient descent training than the error surface of the IIR M LP.
File Format	PDF HTM / HTML
Alternate Webpage(s)	http://www.neci.nec.com/~lawrence/papers/gamma-nnsp97/gamma-nnsp97.ps.gz
Alternate Webpage(s)	http://www.neci.nec.com/~giles/papers/NNSP97.gamma.MLP.ps.Z
Alternate Webpage(s)	http://www.neci.nj.nec.com/homepages/lawrence/papers/gamma-nnsp97/gamma-nnsp97-a4.ps.gz
Language	English
Access Restriction	Open
Subject Keyword	Amma Curse of dimensionality Flash memory Gamma correction Gradient descent Infinite impulse response Interferon Type II Large Memory Disorders Memory-level parallelism Phoneme Population Parameter Quad Flat No-leads package Synaptic Package Manager Time delay neural network Weight anatomical layer
Content Type	Text
Resource Type	Article