Stein Variational Gradient Descent as Gradient Flow

Stein Variational Gradient Descent as Gradient Flow. Liu, Q.
abstract bibtex

Stein variational gradient descent (SVGD) is a deterministic sampling algorithm that iteratively transports a set of particles to approximate given distributions, based on a gradient-based update that guarantees to optimally decrease the KL divergence within a function space. This paper develops the ﬁrst theoretical analysis on SVGD. We establish that the empirical measures of the SVGD samples weakly converge to the target distribution, and show that the asymptotic behavior of SVGD is characterized by a nonlinear Fokker-Planck equation known as Vlasov equation in physics. We develop a geometric perspective that views SVGD as a gradient ﬂow of the KL divergence functional under a new metric structure on the space of distributions induced by Stein operator.

@article{liu_stein_nodate,
	title = {Stein {Variational} {Gradient} {Descent} as {Gradient} {Flow}},
	abstract = {Stein variational gradient descent (SVGD) is a deterministic sampling algorithm that iteratively transports a set of particles to approximate given distributions, based on a gradient-based update that guarantees to optimally decrease the KL divergence within a function space. This paper develops the ﬁrst theoretical analysis on SVGD. We establish that the empirical measures of the SVGD samples weakly converge to the target distribution, and show that the asymptotic behavior of SVGD is characterized by a nonlinear Fokker-Planck equation known as Vlasov equation in physics. We develop a geometric perspective that views SVGD as a gradient ﬂow of the KL divergence functional under a new metric structure on the space of distributions induced by Stein operator.},
	language = {en},
	author = {Liu, Qiang},
	keywords = {/unread, ⛔ No DOI found},
	pages = {9},
}

Downloads: 0

{"_id":"2aHLHWuS2DLYjTeGu","bibbaseid":"liu-steinvariationalgradientdescentasgradientflow","author_short":["Liu, Q."],"bibdata":{"bibtype":"article","type":"article","title":"Stein Variational Gradient Descent as Gradient Flow","abstract":"Stein variational gradient descent (SVGD) is a deterministic sampling algorithm that iteratively transports a set of particles to approximate given distributions, based on a gradient-based update that guarantees to optimally decrease the KL divergence within a function space. This paper develops the ﬁrst theoretical analysis on SVGD. We establish that the empirical measures of the SVGD samples weakly converge to the target distribution, and show that the asymptotic behavior of SVGD is characterized by a nonlinear Fokker-Planck equation known as Vlasov equation in physics. We develop a geometric perspective that views SVGD as a gradient ﬂow of the KL divergence functional under a new metric structure on the space of distributions induced by Stein operator.","language":"en","author":[{"propositions":[],"lastnames":["Liu"],"firstnames":["Qiang"],"suffixes":[]}],"keywords":"/unread, ⛔ No DOI found","pages":"9","bibtex":"@article{liu_stein_nodate,\n\ttitle = {Stein {Variational} {Gradient} {Descent} as {Gradient} {Flow}},\n\tabstract = {Stein variational gradient descent (SVGD) is a deterministic sampling algorithm that iteratively transports a set of particles to approximate given distributions, based on a gradient-based update that guarantees to optimally decrease the KL divergence within a function space. This paper develops the ﬁrst theoretical analysis on SVGD. We establish that the empirical measures of the SVGD samples weakly converge to the target distribution, and show that the asymptotic behavior of SVGD is characterized by a nonlinear Fokker-Planck equation known as Vlasov equation in physics. We develop a geometric perspective that views SVGD as a gradient ﬂow of the KL divergence functional under a new metric structure on the space of distributions induced by Stein operator.},\n\tlanguage = {en},\n\tauthor = {Liu, Qiang},\n\tkeywords = {/unread, ⛔ No DOI found},\n\tpages = {9},\n}\n\n","author_short":["Liu, Q."],"key":"liu_stein_nodate","id":"liu_stein_nodate","bibbaseid":"liu-steinvariationalgradientdescentasgradientflow","role":"author","urls":{},"keyword":["/unread","⛔ No DOI found"],"metadata":{"authorlinks":{}},"html":""},"bibtype":"article","biburl":"https://bibbase.org/zotero/victorjhu","dataSources":["CmHEoydhafhbkXXt5"],"keywords":["/unread","⛔ no doi found"],"search_terms":["stein","variational","gradient","descent","gradient","flow","liu"],"title":"Stein Variational Gradient Descent as Gradient Flow","year":null}