Improved learning algorithms for restricted Boltzmann machines

dc.contributorAalto-yliopistofi
dc.contributorAalto Universityen
dc.contributor.advisorIlin, Alexander
dc.contributor.advisorRaiko, Tapani
dc.contributor.authorCho, KyungHyun
dc.contributor.departmentTietotekniikan laitosfi
dc.contributor.supervisorKarhunen, Juha
dc.date.accessioned2012-07-02T08:08:54Z
dc.date.available2012-07-02T08:08:54Z
dc.date.issued2011
dc.description.abstractA restricted Boltzmann machine (RBM) is often used as a building block for constructing deep neural networks and deep generative models which have gained popularity recently as one way to learn complex and large probabilistic models. In these deep models, it is generally known that the layer-wise pretraining of RBMs facilitates finding a more accurate model for the data. It is, hence, important to have an efficient learning method for RBM. The conventional learning is mostly performed using the stochastic gradients, often, with the approximate method such as contrastive divergence (CD) learning to overcome the computational difficulty. Unfortunately, training RBMs with this approach is known to be difficult, as learning easily diverges after initial convergence. This difficulty has been reported recently by many researchers. This thesis contributes important improvements that address the difficulty of training RBMs. Based on an advanced Markov-Chain Monte-Carlo sampling method called parallel tempering (PT), the thesis proposes a PT learning which can replace CD learning. In terms of both the learning performance and the computational overhead, PT learning is shown to be superior to CD learning through various experiments. The thesis also tackles the problem of choosing the right learning parameter by proposing a new algorithm, the adaptive learning rate, which is able to automatically choose the right learning rate during learning. A closer observation into the update rules suggested that learning by the traditional update rules is easily distracted depending on the representation of data sets. Based on this observation, the thesis proposes a new set of gradient update rules that are more robust to the representation of training data sets and the learning parameters. Extensive experiments on various data sets confirmed that the proposed rules indeed improve learning significantly. Additionally, a Gaussian-Bernoulli RBM (GBRBM) which is a variant of an RBM that can learn continuous real-valued data sets is reviewed, and the proposed improvements are tested upon it. The experiments showed that the improvements could also be made for GBRBMs.en
dc.format.extentxii + 84
dc.format.mimetypeapplication/pdf
dc.identifier.urihttps://aaltodoc.aalto.fi/handle/123456789/3666
dc.identifier.urnURN:NBN:fi:aalto-201207022632
dc.language.isoenen
dc.programme.majorInformaatiotekniikkafi
dc.programme.mcodeT-61
dc.publisherAalto-yliopistofi
dc.publisherAalto Universityen
dc.rights.accesslevelopenAccess
dc.subject.keywordBoltzmann machineen
dc.subject.keywordrestricted Boltzmann machineen
dc.subject.keywordannealed importance samplingen
dc.subject.keywordparaller temperingen
dc.subject.keywordenhanced gradienten
dc.subject.keywordadaptive learning rateen
dc.subject.keywordGaussian-Bernoulli restricted Boltzmann machineen
dc.subject.keyworddeep learningen
dc.titleImproved learning algorithms for restricted Boltzmann machinesen
dc.typeG2 Pro gradu, diplomityöfi
dc.type.dcmitypetexten
dc.type.okmG2 Pro gradu, diplomityö
dc.type.ontasotDiplomityöfi
dc.type.ontasotMaster's thesisen
dc.type.publicationmasterThesis
local.aalto.digifolderAalto_09562
local.aalto.idinssi41637
local.aalto.openaccessyes
Files
Original bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
urn100427.pdf
Size:
5.94 MB
Format:
Adobe Portable Document Format