The Kernel Gibbs Sampler

The Kernel Gibbs Sampler. Graepel, T. & Herbrich, R. In Advances in Neural Information Processing Systems 13, pages 514--520, Denver, 2000. The MIT Press.
abstract bibtex

We present an algorithm that samples the hypothesis space of kernel classifiers. Given a uniform prior over normalised weight vectors and a likelihood based on a model of label noise leads to a piecewise constant posterior that can be sampled by the kernel Gibbs sampler (KGS). The KGS is a Markov Chain Monte Carlo method that chooses a random direction in parameter space and samples from the resulting piecewise constant density along the line chosen. The KGS can be used as an analytical tool for the exploration of Bayesian transduction, Bayes point machines, active learning, and evidence-based model selection on small data sets that are contaminated with label noise. For a simple toy example we demonstrate experimentally how a Bayes point machine based on the KGS outperforms an SVM that is incapable of taking into account label noise.

@inproceedings{DBLP:conf/nips/GraepelH00,
abstract = {We present an algorithm that samples the hypothesis space of kernel classifiers. Given a uniform prior over normalised weight vectors and a likelihood based on a model of label noise leads to a piecewise constant posterior that can be sampled by the kernel Gibbs sampler (KGS). The KGS is a Markov Chain Monte Carlo method that chooses a random direction in parameter space and samples from the resulting piecewise constant density along the line chosen. The KGS can be used as an analytical tool for the exploration of Bayesian transduction, Bayes point machines, active learning, and evidence-based model selection on small data sets that are contaminated with label noise. For a simple toy example we demonstrate experimentally how a Bayes point machine based on the KGS outperforms an SVM that is incapable of taking into account label noise.},
address = {Denver},
author = {Graepel, Thore and Herbrich, Ralf},
booktitle = {Advances in Neural Information Processing Systems 13},
file = {:Users/rherb/Dropbox/Documents/tex/nips2000/kgibbs/kgibbs.pdf:pdf},
pages = {514--520},
publisher = {The MIT Press},
title = {{The Kernel Gibbs Sampler}},
year = {2000}
}

Downloads: 0

{"_id":{"_str":"53421b61ecd21cdc070003e3"},"__v":9,"authorIDs":["5456e9a38b01c8193000005e","54576c282abc8e9f370003ae"],"author_short":["Graepel, T.","Herbrich, R."],"bibbaseid":"graepel-herbrich-thekernelgibbssampler-2000","bibdata":{"bibtype":"inproceedings","type":"inproceedings","abstract":"We present an algorithm that samples the hypothesis space of kernel classifiers. Given a uniform prior over normalised weight vectors and a likelihood based on a model of label noise leads to a piecewise constant posterior that can be sampled by the kernel Gibbs sampler (KGS). The KGS is a Markov Chain Monte Carlo method that chooses a random direction in parameter space and samples from the resulting piecewise constant density along the line chosen. The KGS can be used as an analytical tool for the exploration of Bayesian transduction, Bayes point machines, active learning, and evidence-based model selection on small data sets that are contaminated with label noise. For a simple toy example we demonstrate experimentally how a Bayes point machine based on the KGS outperforms an SVM that is incapable of taking into account label noise.","address":"Denver","author":[{"propositions":[],"lastnames":["Graepel"],"firstnames":["Thore"],"suffixes":[]},{"propositions":[],"lastnames":["Herbrich"],"firstnames":["Ralf"],"suffixes":[]}],"booktitle":"Advances in Neural Information Processing Systems 13","file":":Users/rherb/Dropbox/Documents/tex/nips2000/kgibbs/kgibbs.pdf:pdf","pages":"514--520","publisher":"The MIT Press","title":"The Kernel Gibbs Sampler","year":"2000","bibtex":"@inproceedings{DBLP:conf/nips/GraepelH00,\nabstract = {We present an algorithm that samples the hypothesis space of kernel classifiers. Given a uniform prior over normalised weight vectors and a likelihood based on a model of label noise leads to a piecewise constant posterior that can be sampled by the kernel Gibbs sampler (KGS). The KGS is a Markov Chain Monte Carlo method that chooses a random direction in parameter space and samples from the resulting piecewise constant density along the line chosen. The KGS can be used as an analytical tool for the exploration of Bayesian transduction, Bayes point machines, active learning, and evidence-based model selection on small data sets that are contaminated with label noise. For a simple toy example we demonstrate experimentally how a Bayes point machine based on the KGS outperforms an SVM that is incapable of taking into account label noise.},\naddress = {Denver},\nauthor = {Graepel, Thore and Herbrich, Ralf},\nbooktitle = {Advances in Neural Information Processing Systems 13},\nfile = {:Users/rherb/Dropbox/Documents/tex/nips2000/kgibbs/kgibbs.pdf:pdf},\npages = {514--520},\npublisher = {The MIT Press},\ntitle = {{The Kernel Gibbs Sampler}},\nyear = {2000}\n}\n","author_short":["Graepel, T.","Herbrich, R."],"key":"DBLP:conf/nips/GraepelH00","id":"DBLP:conf/nips/GraepelH00","bibbaseid":"graepel-herbrich-thekernelgibbssampler-2000","role":"author","urls":{},"downloads":0,"html":""},"bibtype":"inproceedings","biburl":"http://herbrich.me/bib/herbrich.bib","downloads":0,"keywords":[],"search_terms":["kernel","gibbs","sampler","graepel","herbrich"],"title":"The Kernel Gibbs Sampler","year":2000,"dataSources":["y2DvMgAcqeDpXQ6ds"]}