Matches in SemOpenAlex for { <https://semopenalex.org/work/W4313596422> ?p ?o ?g. }
Showing items 1 to 83 of
83
with 100 items per page.
- W4313596422 endingPage "17443" @default.
- W4313596422 startingPage "17429" @default.
- W4313596422 abstract "Deep neural networks exploiting million parameters are currently the norm. This is a potential issue because of the great number of computations needed for training, and the possible loss of generalization performance of overparameterized networks. We propose in this paper a method for learning sparse neural topologies via a regularization approach that identifies nonrelevant weights in any type of layer (i.e., convolutional, fully connected, attention and embedding ones) and selectively shrinks their norm while performing a standard back-propagation update for relevant layers. This technique, which is an improvement of classical weight decay, is based on the definition of a regularization term that can be added to any loss function regardless of its form, resulting in a unified general framework exploitable in many different contexts. The actual elimination of parameters identified as irrelevant is handled by an iterative pruning algorithm. To explore the possibility of an interdisciplinary use of our proposed technique, we test it on six different image classification and natural language generation tasks, among which four are based on real datasets. We reach state-of-the-art performance in one out of four imaging tasks while obtaining results better than competitors for the others and one out of two of the considered language generation tasks, both in terms of compression and metrics." @default.
- W4313596422 created "2023-01-06" @default.
- W4313596422 creator A5004307208 @default.
- W4313596422 creator A5015514303 @default.
- W4313596422 creator A5049044181 @default.
- W4313596422 date "2023-01-05" @default.
- W4313596422 modified "2023-09-27" @default.
- W4313596422 title "Regularization-based pruning of irrelevant weights in deep neural architectures" @default.
- W4313596422 cites W2081580037 @default.
- W4313596422 cites W2097117768 @default.
- W4313596422 cites W2101105183 @default.
- W4313596422 cites W2108598243 @default.
- W4313596422 cites W2145607950 @default.
- W4313596422 cites W2147800946 @default.
- W4313596422 cites W2194775991 @default.
- W4313596422 cites W2257408573 @default.
- W4313596422 cites W2605177561 @default.
- W4313596422 cites W2807961551 @default.
- W4313596422 cites W2914397182 @default.
- W4313596422 cites W2946794439 @default.
- W4313596422 cites W2963223345 @default.
- W4313596422 cites W2963382930 @default.
- W4313596422 cites W2963592583 @default.
- W4313596422 cites W2963743213 @default.
- W4313596422 cites W2963855133 @default.
- W4313596422 cites W2970958999 @default.
- W4313596422 cites W2985067290 @default.
- W4313596422 cites W2991849637 @default.
- W4313596422 cites W3033960544 @default.
- W4313596422 cites W3034513523 @default.
- W4313596422 cites W3035160838 @default.
- W4313596422 cites W3035302306 @default.
- W4313596422 cites W3048709363 @default.
- W4313596422 cites W3100837220 @default.
- W4313596422 cites W3203376206 @default.
- W4313596422 doi "https://doi.org/10.1007/s10489-022-04353-y" @default.
- W4313596422 hasPublicationYear "2023" @default.
- W4313596422 type Work @default.
- W4313596422 citedByCount "0" @default.
- W4313596422 crossrefType "journal-article" @default.
- W4313596422 hasAuthorship W4313596422A5004307208 @default.
- W4313596422 hasAuthorship W4313596422A5015514303 @default.
- W4313596422 hasAuthorship W4313596422A5049044181 @default.
- W4313596422 hasBestOaLocation W43135964222 @default.
- W4313596422 hasConcept C11413529 @default.
- W4313596422 hasConcept C119857082 @default.
- W4313596422 hasConcept C154945302 @default.
- W4313596422 hasConcept C2776135515 @default.
- W4313596422 hasConcept C41008148 @default.
- W4313596422 hasConcept C41608201 @default.
- W4313596422 hasConcept C45374587 @default.
- W4313596422 hasConcept C50644808 @default.
- W4313596422 hasConcept C81363708 @default.
- W4313596422 hasConceptScore W4313596422C11413529 @default.
- W4313596422 hasConceptScore W4313596422C119857082 @default.
- W4313596422 hasConceptScore W4313596422C154945302 @default.
- W4313596422 hasConceptScore W4313596422C2776135515 @default.
- W4313596422 hasConceptScore W4313596422C41008148 @default.
- W4313596422 hasConceptScore W4313596422C41608201 @default.
- W4313596422 hasConceptScore W4313596422C45374587 @default.
- W4313596422 hasConceptScore W4313596422C50644808 @default.
- W4313596422 hasConceptScore W4313596422C81363708 @default.
- W4313596422 hasIssue "14" @default.
- W4313596422 hasLocation W43135964221 @default.
- W4313596422 hasLocation W43135964222 @default.
- W4313596422 hasOpenAccess W4313596422 @default.
- W4313596422 hasPrimaryLocation W43135964221 @default.
- W4313596422 hasRelatedWork W2337926734 @default.
- W4313596422 hasRelatedWork W3027997911 @default.
- W4313596422 hasRelatedWork W4225307033 @default.
- W4313596422 hasRelatedWork W4287776258 @default.
- W4313596422 hasRelatedWork W4312501200 @default.
- W4313596422 hasRelatedWork W4313050734 @default.
- W4313596422 hasRelatedWork W4320802194 @default.
- W4313596422 hasRelatedWork W4366224123 @default.
- W4313596422 hasRelatedWork W4381832759 @default.
- W4313596422 hasRelatedWork W1629725936 @default.
- W4313596422 hasVolume "53" @default.
- W4313596422 isParatext "false" @default.
- W4313596422 isRetracted "false" @default.
- W4313596422 workType "article" @default.