FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3203385474> ?p ?o ?g. }

• W3203385474 abstract "A “bigger is better” explosion in the number of parameters in deep neural networks has made it increasingly challenging to make state-of-the-art networks accessible in compute-restricted environments. Compression techniques have taken on renewed importance as a way to bridge the gap. However, evaluation of the trade-offs incurred by popular compression techniques has been centered on high-resource datasets. In this work, we instead consider the impact of compression in a data-limited regime. We introduce the term low-resource double bind to refer to the co-occurrence of data limitations and compute resource constraints. This is a common setting for NLP for low-resource languages, yet the trade-offs in performance are poorly studied. Our work offers surprising insights into the relationship between capacity and generalization in data-limited regimes for the task of machine translation. Our experiments on magnitude pruning for translations from English into Yoruba, Hausa, Igbo and German show that in low-resource regimes, sparsity preserves performance on frequent sentences but has a disparate impact on infrequent ones. However, it improves robustness to out-of-distribution shifts, especially for datasets that are very distinct from the training distribution. Our findings suggest that sparsity can play a beneficial role at curbing memorization of low frequency attributes, and therefore offers a promising solution to the low-resource double bind." @default.
• W3203385474 created "2021-10-11" @default.
• W3203385474 creator A5045507345 @default.
• W3203385474 creator A5048307591 @default.
• W3203385474 creator A5078850040 @default.
• W3203385474 date "2021-01-01" @default.
• W3203385474 modified "2023-10-18" @default.
• W3203385474 title "The Low-Resource Double Bind: An Empirical Study of Pruning for Low-Resource Machine Translation" @default.
• W3203385474 cites W133229983 @default.
• W3203385474 cites W1570197553 @default.
• W3203385474 cites W1594170634 @default.
• W3203385474 cites W1821462560 @default.
• W3203385474 cites W1825672851 @default.
• W3203385474 cites W1999085092 @default.
• W3203385474 cites W2053782355 @default.
• W3203385474 cites W2101105183 @default.
• W3203385474 cites W2114766824 @default.
• W3203385474 cites W2125389748 @default.
• W3203385474 cites W2145085734 @default.
• W3203385474 cites W2148365102 @default.
• W3203385474 cites W2156150815 @default.
• W3203385474 cites W2279098554 @default.
• W3203385474 cites W2524428287 @default.
• W3203385474 cites W2573818331 @default.
• W3203385474 cites W2581082771 @default.
• W3203385474 cites W2608554408 @default.
• W3203385474 cites W2612445135 @default.
• W3203385474 cites W2732951378 @default.
• W3203385474 cites W2758162718 @default.
• W3203385474 cites W2764043458 @default.
• W3203385474 cites W2771655537 @default.
• W3203385474 cites W2796345789 @default.
• W3203385474 cites W2804309962 @default.
• W3203385474 cites W2915589364 @default.
• W3203385474 cites W2948223045 @default.
• W3203385474 cites W2949303037 @default.
• W3203385474 cites W2951165112 @default.
• W3203385474 cites W2951451051 @default.
• W3203385474 cites W2962784628 @default.
• W3203385474 cites W2962944188 @default.
• W3203385474 cites W2963000224 @default.
• W3203385474 cites W2963088995 @default.
• W3203385474 cites W2963122961 @default.
• W3203385474 cites W2963374099 @default.
• W3203385474 cites W2963403868 @default.
• W3203385474 cites W2963418779 @default.
• W3203385474 cites W2963500086 @default.
• W3203385474 cites W2963532001 @default.
• W3203385474 cites W2963643655 @default.
• W3203385474 cites W2963674932 @default.
• W3203385474 cites W2963981420 @default.
• W3203385474 cites W2980186997 @default.
• W3203385474 cites W2986700214 @default.
• W3203385474 cites W2986915867 @default.
• W3203385474 cites W2990844796 @default.
• W3203385474 cites W3010156474 @default.
• W3203385474 cites W3013741422 @default.
• W3203385474 cites W3030384960 @default.
• W3203385474 cites W3034340181 @default.
• W3203385474 cites W3034716087 @default.
• W3203385474 cites W3035032094 @default.
• W3203385474 cites W3041148953 @default.
• W3203385474 cites W3081141044 @default.
• W3203385474 cites W3091818438 @default.
• W3203385474 cites W3092348244 @default.
• W3203385474 cites W3098341425 @default.
• W3203385474 cites W3100352836 @default.
• W3203385474 cites W3100501376 @default.
• W3203385474 cites W3100684858 @default.
• W3203385474 cites W3101493110 @default.
• W3203385474 cites W3101584733 @default.
• W3203385474 cites W3101731278 @default.
• W3203385474 cites W3114304470 @default.
• W3203385474 cites W3120519792 @default.
• W3203385474 cites W3126203713 @default.
• W3203385474 cites W3130660608 @default.
• W3203385474 cites W3136219906 @default.
• W3203385474 cites W3136598139 @default.
• W3203385474 cites W3139460307 @default.
• W3203385474 cites W3155808134 @default.
• W3203385474 cites W3156891177 @default.
• W3203385474 cites W3172669006 @default.
• W3203385474 cites W3173417753 @default.
• W3203385474 cites W3102138045 @default.
• W3203385474 doi "https://doi.org/10.18653/v1/2021.findings-emnlp.282" @default.
• W3203385474 hasPublicationYear "2021" @default.
• W3203385474 type Work @default.
• W3203385474 sameAs 3203385474 @default.
• W3203385474 citedByCount "2" @default.
• W3203385474 countsByYear W32033854742023 @default.
• W3203385474 crossrefType "proceedings-article" @default.
• W3203385474 hasAuthorship W3203385474A5045507345 @default.
• W3203385474 hasAuthorship W3203385474A5048307591 @default.
• W3203385474 hasAuthorship W3203385474A5078850040 @default.
• W3203385474 hasBestOaLocation W32033854741 @default.
• W3203385474 hasConcept C104317684 @default.
• W3203385474 hasConcept C108010975 @default.
• W3203385474 hasConcept C119857082 @default.
• W3203385474 hasConcept C154945302 @default.
• W3203385474 hasConcept C185592680 @default.

• next