FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3195845308> ?p ?o ?g. }

• W3195845308 abstract "While deep learning has been very beneficial in data-rich settings, tasks with smaller training set often resort to pre-training or multitask learning to leverage data from other tasks. In this case, careful consideration is needed to select tasks and model parameterizations such that updates from the auxiliary tasks actually help the primary task. We seek to alleviate this burden by formulating a model-agnostic framework that performs fine-grained manipulation of the auxiliary task gradients. We propose to decompose auxiliary updates into directions which help, damage or leave the primary task loss unchanged. This allows weighting the update directions differently depending on their impact on the problem of interest. We present a novel and efficient algorithm for that purpose and show its advantage in practice. Our method leverages efficient automatic differentiation procedures and randomized singular value decomposition for scalability. We show that our framework is generic and encompasses some prior work as particular cases. Our approach consistently outperforms strong and widely used baselines when leveraging out-of-distribution data for Text and Image classification tasks." @default.
• W3195845308 created "2021-08-30" @default.
• W3195845308 creator A5020785659 @default.
• W3195845308 creator A5041114293 @default.
• W3195845308 creator A5065912572 @default.
• W3195845308 date "2021-08-25" @default.
• W3195845308 modified "2023-09-27" @default.
• W3195845308 title "Auxiliary Task Update Decomposition: The Good, The Bad and The Neutral" @default.
• W3195845308 cites W1815076433 @default.
• W3195845308 cites W1853291804 @default.
• W3195845308 cites W1896424170 @default.
• W3195845308 cites W1905522558 @default.
• W3195845308 cites W2006903949 @default.
• W3195845308 cites W2027731328 @default.
• W3195845308 cites W2040387238 @default.
• W3195845308 cites W2108598243 @default.
• W3195845308 cites W2117130368 @default.
• W3195845308 cites W2117278770 @default.
• W3195845308 cites W2117756735 @default.
• W3195845308 cites W2194775991 @default.
• W3195845308 cites W2401231614 @default.
• W3195845308 cites W2624871570 @default.
• W3195845308 cites W2654090051 @default.
• W3195845308 cites W2804935296 @default.
• W3195845308 cites W2883725317 @default.
• W3195845308 cites W2890538051 @default.
• W3195845308 cites W2891076394 @default.
• W3195845308 cites W2899771611 @default.
• W3195845308 cites W2901026139 @default.
• W3195845308 cites W2902347140 @default.
• W3195845308 cites W2913340405 @default.
• W3195845308 cites W2944407464 @default.
• W3195845308 cites W2948666538 @default.
• W3195845308 cites W2952468927 @default.
• W3195845308 cites W2952677972 @default.
• W3195845308 cites W2962724315 @default.
• W3195845308 cites W2962743139 @default.
• W3195845308 cites W2963341956 @default.
• W3195845308 cites W2963393838 @default.
• W3195845308 cites W2963466845 @default.
• W3195845308 cites W2963877604 @default.
• W3195845308 cites W2964121744 @default.
• W3195845308 cites W2965373594 @default.
• W3195845308 cites W2970597249 @default.
• W3195845308 cites W2970854840 @default.
• W3195845308 cites W2987741655 @default.
• W3195845308 cites W2991420892 @default.
• W3195845308 cites W2996490626 @default.
• W3195845308 cites W2997359900 @default.
• W3195845308 cites W3034214887 @default.
• W3195845308 cites W3034238904 @default.
• W3195845308 cites W3035608860 @default.
• W3195845308 cites W3037967334 @default.
• W3195845308 cites W3118608800 @default.
• W3195845308 cites W2551176409 @default.
• W3195845308 cites W2809584662 @default.
• W3195845308 hasPublicationYear "2021" @default.
• W3195845308 type Work @default.
• W3195845308 sameAs 3195845308 @default.
• W3195845308 citedByCount "0" @default.
• W3195845308 crossrefType "posted-content" @default.
• W3195845308 hasAuthorship W3195845308A5020785659 @default.
• W3195845308 hasAuthorship W3195845308A5041114293 @default.
• W3195845308 hasAuthorship W3195845308A5065912572 @default.
• W3195845308 hasConcept C119857082 @default.
• W3195845308 hasConcept C124101348 @default.
• W3195845308 hasConcept C124681953 @default.
• W3195845308 hasConcept C126838900 @default.
• W3195845308 hasConcept C153083717 @default.
• W3195845308 hasConcept C154945302 @default.
• W3195845308 hasConcept C162324750 @default.
• W3195845308 hasConcept C177264268 @default.
• W3195845308 hasConcept C183115368 @default.
• W3195845308 hasConcept C187736073 @default.
• W3195845308 hasConcept C18903297 @default.
• W3195845308 hasConcept C199360897 @default.
• W3195845308 hasConcept C22789450 @default.
• W3195845308 hasConcept C2780451532 @default.
• W3195845308 hasConcept C28006648 @default.
• W3195845308 hasConcept C41008148 @default.
• W3195845308 hasConcept C48044578 @default.
• W3195845308 hasConcept C71924100 @default.
• W3195845308 hasConcept C77088390 @default.
• W3195845308 hasConcept C86803240 @default.
• W3195845308 hasConceptScore W3195845308C119857082 @default.
• W3195845308 hasConceptScore W3195845308C124101348 @default.
• W3195845308 hasConceptScore W3195845308C124681953 @default.
• W3195845308 hasConceptScore W3195845308C126838900 @default.
• W3195845308 hasConceptScore W3195845308C153083717 @default.
• W3195845308 hasConceptScore W3195845308C154945302 @default.
• W3195845308 hasConceptScore W3195845308C162324750 @default.
• W3195845308 hasConceptScore W3195845308C177264268 @default.
• W3195845308 hasConceptScore W3195845308C183115368 @default.
• W3195845308 hasConceptScore W3195845308C187736073 @default.
• W3195845308 hasConceptScore W3195845308C18903297 @default.
• W3195845308 hasConceptScore W3195845308C199360897 @default.
• W3195845308 hasConceptScore W3195845308C22789450 @default.
• W3195845308 hasConceptScore W3195845308C2780451532 @default.
• W3195845308 hasConceptScore W3195845308C28006648 @default.
• W3195845308 hasConceptScore W3195845308C41008148 @default.

• next