Matches in SemOpenAlex for { <https://semopenalex.org/work/W2894337195> ?p ?o ?g. }
Showing items 1 to 100 of
100
with 100 items per page.
- W2894337195 abstract "We combine two recent lines of research on sublinear clustering to significantly increase the efficiency in the training of large-scale Gaussian mixture models (GMMs). First, we use a novel truncated variational EM approach for GMMs with isotropic Gaussians in order to increase clustering efficiency for large $C$ (many clusters). Second, we use recent coreset approaches to increase clustering efficiency for large $N$ (many data points). In order to derive a novel accelerated algorithm, we first show analytically how variational EM and coreset objectives can be merged to give rise to a new, combined clustering objective. Each iteration of the novel algorithm derived from this merged objective is then shown to have a run-time cost of $mathcal{O}(N' G^2 D)$ per iteration, where $N'<N$ is the coreset size and $G^2<C$ is a constant related to the extent of local cluster neighborhoods. While the approach strongly reduces the number of distance evaluations per EM iteration, we observe the iterations to maintain a very effective increase of the clustering objective. In a series of numerical experiments, we use efficient seeding for initialization and measure the net computational demand of the merged approach in comparison to other recent approaches. For standard benchmarks which evaluate the trade-off between values of the clustering objective and clustering efficiency, the merged approach significantly improves the state-of-the-art. Depending on the data set and number of clusters, we observe several times (and up to an order of magnitude) faster execution times to reach the same quantization errors compared to the best recent approaches such as highly efficient coreset-based $k$-means." @default.
- W2894337195 created "2018-10-05" @default.
- W2894337195 creator A5022342273 @default.
- W2894337195 creator A5023612311 @default.
- W2894337195 creator A5031051637 @default.
- W2894337195 date "2018-10-01" @default.
- W2894337195 modified "2023-09-27" @default.
- W2894337195 title "Accelerated Training of Large-Scale Gaussian Mixtures by a Merger of Sublinear Approaches." @default.
- W2894337195 cites W118481696 @default.
- W2894337195 cites W1530232144 @default.
- W2894337195 cites W1556219185 @default.
- W2894337195 cites W1622263187 @default.
- W2894337195 cites W1999668761 @default.
- W2894337195 cites W2045964207 @default.
- W2894337195 cites W2048305092 @default.
- W2894337195 cites W2076257979 @default.
- W2894337195 cites W2118858186 @default.
- W2894337195 cites W2125621954 @default.
- W2894337195 cites W2141522908 @default.
- W2894337195 cites W2146200992 @default.
- W2894337195 cites W2166851633 @default.
- W2894337195 cites W2510191442 @default.
- W2894337195 cites W2550969209 @default.
- W2894337195 cites W2552050672 @default.
- W2894337195 cites W2567948266 @default.
- W2894337195 cites W2601251344 @default.
- W2894337195 cites W2623109347 @default.
- W2894337195 cites W2807006342 @default.
- W2894337195 cites W2963264219 @default.
- W2894337195 cites W2963555385 @default.
- W2894337195 cites W2963873208 @default.
- W2894337195 cites W2964157900 @default.
- W2894337195 cites W3102268593 @default.
- W2894337195 hasPublicationYear "2018" @default.
- W2894337195 type Work @default.
- W2894337195 sameAs 2894337195 @default.
- W2894337195 citedByCount "0" @default.
- W2894337195 crossrefType "posted-content" @default.
- W2894337195 hasAuthorship W2894337195A5022342273 @default.
- W2894337195 hasAuthorship W2894337195A5023612311 @default.
- W2894337195 hasAuthorship W2894337195A5031051637 @default.
- W2894337195 hasConcept C11413529 @default.
- W2894337195 hasConcept C114466953 @default.
- W2894337195 hasConcept C114614502 @default.
- W2894337195 hasConcept C117160843 @default.
- W2894337195 hasConcept C121332964 @default.
- W2894337195 hasConcept C126255220 @default.
- W2894337195 hasConcept C154945302 @default.
- W2894337195 hasConcept C163716315 @default.
- W2894337195 hasConcept C177264268 @default.
- W2894337195 hasConcept C199360897 @default.
- W2894337195 hasConcept C33923547 @default.
- W2894337195 hasConcept C41008148 @default.
- W2894337195 hasConcept C61224824 @default.
- W2894337195 hasConcept C62520636 @default.
- W2894337195 hasConcept C73555534 @default.
- W2894337195 hasConcept C94641424 @default.
- W2894337195 hasConceptScore W2894337195C11413529 @default.
- W2894337195 hasConceptScore W2894337195C114466953 @default.
- W2894337195 hasConceptScore W2894337195C114614502 @default.
- W2894337195 hasConceptScore W2894337195C117160843 @default.
- W2894337195 hasConceptScore W2894337195C121332964 @default.
- W2894337195 hasConceptScore W2894337195C126255220 @default.
- W2894337195 hasConceptScore W2894337195C154945302 @default.
- W2894337195 hasConceptScore W2894337195C163716315 @default.
- W2894337195 hasConceptScore W2894337195C177264268 @default.
- W2894337195 hasConceptScore W2894337195C199360897 @default.
- W2894337195 hasConceptScore W2894337195C33923547 @default.
- W2894337195 hasConceptScore W2894337195C41008148 @default.
- W2894337195 hasConceptScore W2894337195C61224824 @default.
- W2894337195 hasConceptScore W2894337195C62520636 @default.
- W2894337195 hasConceptScore W2894337195C73555534 @default.
- W2894337195 hasConceptScore W2894337195C94641424 @default.
- W2894337195 hasLocation W28943371951 @default.
- W2894337195 hasOpenAccess W2894337195 @default.
- W2894337195 hasPrimaryLocation W28943371951 @default.
- W2894337195 hasRelatedWork W1495441429 @default.
- W2894337195 hasRelatedWork W1546145162 @default.
- W2894337195 hasRelatedWork W2032582793 @default.
- W2894337195 hasRelatedWork W2036116575 @default.
- W2894337195 hasRelatedWork W2109426916 @default.
- W2894337195 hasRelatedWork W2123513914 @default.
- W2894337195 hasRelatedWork W2174057495 @default.
- W2894337195 hasRelatedWork W2307341269 @default.
- W2894337195 hasRelatedWork W2510191442 @default.
- W2894337195 hasRelatedWork W2579685248 @default.
- W2894337195 hasRelatedWork W2739634688 @default.
- W2894337195 hasRelatedWork W2890596636 @default.
- W2894337195 hasRelatedWork W2911597709 @default.
- W2894337195 hasRelatedWork W2918668824 @default.
- W2894337195 hasRelatedWork W2948910962 @default.
- W2894337195 hasRelatedWork W2970121096 @default.
- W2894337195 hasRelatedWork W3083518266 @default.
- W2894337195 hasRelatedWork W3119491467 @default.
- W2894337195 hasRelatedWork W3129536844 @default.
- W2894337195 hasRelatedWork W3202618506 @default.
- W2894337195 isParatext "false" @default.
- W2894337195 isRetracted "false" @default.
- W2894337195 magId "2894337195" @default.
- W2894337195 workType "article" @default.