Automatically assigned DDC number: 00632

Manually assigned DDC number: 00632

Number of references: 4

Title: Regularization With a Pruning Prior

Author:

Author:

Subject: Cyril Goutte,Lars Kai Hansen Regularization With a Pruning Prior

Description: We investigate the use of a regularization prior that we show has pruning properties. Analyses are conducted both using a Bayesian framework and with the generalization method, on a simple toy problem. Results are thoroughly compared with those obtained with a traditional weight decay. Keywords: Regularization, Generalization method, Bayesian learning, Evidence framework, Laplace prior, Comparison with weight decay. 1 Gauss and Laplace priors Regularization by weight decay, and the associated Bayesian interpretation involving a Gaussian prior is a well established part of modern neural computation, see e.g., [KH91]. In [HR94] it was shown that if the weight decay parameter is determined by data, either using MacKay's Evidence procedure, or by minimizing generalization error, pruning may result. In a recent communication, [Wil95] elucidated the interesting properties of the Laplacian prior, in particular he showed that this prior directly leads to pruning of small weights, for any fixe...

Contributor: The Pennsylvania State University CiteSeer Archives

Publisher: unknown

Date: 1996-02-14

Pubyear: 1997

Format: ps

Identifier: http://citeseer.ist.psu.edu/150143.html

Source: http://www-poleia.lip6.fr/CONNEX/Articles/goutte-regularization.ps.gz

Language: en

Relation:

Relation:

Relation:

Relation:

Rights: unrestricted

Graph

<?xml   version="1.0"   encoding="UTF-8"?>

<references_metadata>

      <rec   ID="/13312.html"   Type="incollection"   CiteSeer_Book="Maximum   Entropy   and   Bayesian   Methods   in   Applied   Statistics"   CiteSeer_Volume=""   Title="Bayesian   methods:   {G}eneral   Background,"   />

      <rec   ID="/71167.html"   Type="inproceedings"   CiteSeer_Book="Advances   in   Neural   Information   Processing   Systems"   CiteSeer_Volume=""   Title="A   Simple   Weight   Decay   Can   Improve   Generalization,">

            <identifier   Org="ISBN:0198524226"   Paper_ID="/71167.html"   Extracted="0198524226"   DDC="519.5/0285/632"   Normalized_DDC="51950285632"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:0262122413"   Paper_ID="/71167.html"   Extracted="0262122413"   />

            <identifier   Org="ISBN:0262201046"   Paper_ID="/71167.html"   Extracted="0262201046"   />

            <identifier   Org="ISBN:0262661160"   Paper_ID="/71167.html"   Extracted="0262661160"   DDC="153"   Normalized_DDC="153"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:0780312007"   Paper_ID="/71167.html"   Extracted="0780312007"   DDC="006.3"   Normalized_DDC="0063"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:0780319028"   Paper_ID="/71167.html"   Extracted="0780319028"   DDC="006.3"   Normalized_DDC="0063"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:0780332113"   Paper_ID="/71167.html"   Extracted="0780332113"   DDC="006.3/2"   Normalized_DDC="00632"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:0780341236"   Paper_ID="/71167.html"   Extracted="0780341236"   DDC="006.3/2"   Normalized_DDC="00632"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:0805826084"   Paper_ID="/71167.html"   Extracted="0805826084"   />

            <identifier   Org="ISBN:0805835814"   Paper_ID="/71167.html"   Extracted="0805835814"   DDC="153"   Normalized_DDC="153"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:1586032801"   Paper_ID="/71167.html"   Extracted="1586032801"   DDC="006.3"   Normalized_DDC="0063"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:1586033247"   Paper_ID="/71167.html"   Extracted="1586033247"   DDC="006.3/2"   Normalized_DDC="00632"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:1586036211"   Paper_ID="/71167.html"   Extracted="1586036211"   DDC="551.483"   Normalized_DDC="551483"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:1860947581"   Paper_ID="/71167.html"   Extracted="1860947581"   DDC="006.3/2"   Normalized_DDC="00632"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:3540256776"   Paper_ID="/71167.html"   Extracted="3540256776"   DDC="001.01/2"   Normalized_DDC="001012"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:3540262083"   Paper_ID="/71167.html"   Extracted="3540262083"   DDC="006.3/2"   Normalized_DDC="00632"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:4274905357"   Paper_ID="/71167.html"   Extracted="4274905357"   DDC="006.3"   Normalized_DDC="0063"   Normalized_Weight="0.07142857142857142"   />

            <identifier   Org="ISBN:8173713936"   Paper_ID="/71167.html"   Extracted="8173713936"   />

            <identifier   Org="ISBN:9814021237"   Paper_ID="/71167.html"   Extracted="9814021237"   />

      </rec>

      <rec   ID="/434739.html"   Type="article"   CiteSeer_Book="Neural   Computation"   CiteSeer_Volume="4"   Title="{B}ayesian   Interpolation,">

            <identifier   Org="ISBN:1591406714"   Paper_ID="/434739.html"   Extracted="1591406714"   DDC="332.0285/632"   Normalized_DDC="3320285632"   Normalized_Weight="1.0"   />

      </rec>

      <rec   ID="/431588.html"   Type="inproceedings"   CiteSeer_Book="Advances   in   Neural   Information   Processing   Systems   4"   CiteSeer_Volume=""   Title="A   practical   {B}ayesian   framework   for   backprop   networks,"   />

      <rec   ID="SELF"   Type="SELF"   CiteSeer_Book="SELF"   CiteSeer_Volume="SELF"   Title="Regularization   With   a   Pruning   Prior">

            <identifier   Org="ISBN:0471741086"   Paper_ID="SELF"   Extracted="0471741086"   DDC="006.32"   Normalized_DDC="00632"   Normalized_Weight="0.3333333333333333"   />

            <identifier   Org="ISBN:0780335511"   Paper_ID="SELF"   Extracted="0780335511"   />

            <identifier   Org="ISBN:0792382900"   Paper_ID="SELF"   Extracted="0792382900"   DDC="006.3/2"   Normalized_DDC="00632"   Normalized_Weight="0.3333333333333333"   />

            <identifier   Org="ISBN:3540424865"   Paper_ID="SELF"   Extracted="3540424865"   DDC="006.3/2"   Normalized_DDC="00632"   Normalized_Weight="0.3333333333333333"   />

      </rec>

</references_metadata>

www.000webhost.com