Matches in SemOpenAlex for { <https://semopenalex.org/work/W3153940464> ?p ?o ?g. }
- W3153940464 abstract "The two-tower architecture has been widely applied for learning item and user representations, which is important for large-scale recommender systems. Many two-tower models are trained using various in-batch negative sampling strategies, where the effects of such strategies inherently rely on the size of mini-batches. However, training two-tower models with a large batch size is inefficient, as it demands a large volume of memory for item and user contents and consumes a lot of time for feature encoding. Interestingly, we find that neural encoders can output relatively stable features for the same input after warming up in the training process. Based on such facts, we propose a simple yet effective sampling strategy called Cross-Batch Negative Sampling (CBNS), which takes advantage of the encoded item embeddings from recent mini-batches to boost the model training. Both theoretical analysis and empirical evaluations demonstrate the effectiveness and the efficiency of CBNS." @default.
- W3153940464 created "2021-04-26" @default.
- W3153940464 creator A5035927942 @default.
- W3153940464 creator A5048669373 @default.
- W3153940464 creator A5083350101 @default.
- W3153940464 date "2021-07-11" @default.
- W3153940464 modified "2023-10-04" @default.
- W3153940464 title "Cross-Batch Negative Sampling for Training Two-Tower Recommenders" @default.
- W3153940464 cites W2101409192 @default.
- W3153940464 cites W2152808281 @default.
- W3153940464 cites W2158515176 @default.
- W3153940464 cites W2295739661 @default.
- W3153940464 cites W2512971201 @default.
- W3153940464 cites W2640408555 @default.
- W3153940464 cites W2648699835 @default.
- W3153940464 cites W2741249238 @default.
- W3153940464 cites W2798916557 @default.
- W3153940464 cites W2912967843 @default.
- W3153940464 cites W2963085847 @default.
- W3153940464 cites W2963350250 @default.
- W3153940464 cites W2972801466 @default.
- W3153940464 cites W2982108874 @default.
- W3153940464 cites W2982902390 @default.
- W3153940464 cites W2987249037 @default.
- W3153940464 cites W3014828506 @default.
- W3153940464 cites W3035014997 @default.
- W3153940464 cites W3035524453 @default.
- W3153940464 cites W3038033387 @default.
- W3153940464 cites W3080642298 @default.
- W3153940464 cites W3099700870 @default.
- W3153940464 cites W3104748221 @default.
- W3153940464 cites W4297971002 @default.
- W3153940464 doi "https://doi.org/10.1145/3404835.3463032" @default.
- W3153940464 hasPublicationYear "2021" @default.
- W3153940464 type Work @default.
- W3153940464 sameAs 3153940464 @default.
- W3153940464 citedByCount "13" @default.
- W3153940464 countsByYear W31539404642022 @default.
- W3153940464 countsByYear W31539404642023 @default.
- W3153940464 crossrefType "proceedings-article" @default.
- W3153940464 hasAuthorship W3153940464A5035927942 @default.
- W3153940464 hasAuthorship W3153940464A5048669373 @default.
- W3153940464 hasAuthorship W3153940464A5083350101 @default.
- W3153940464 hasBestOaLocation W31539404642 @default.
- W3153940464 hasConcept C106131492 @default.
- W3153940464 hasConcept C111919701 @default.
- W3153940464 hasConcept C118505674 @default.
- W3153940464 hasConcept C119857082 @default.
- W3153940464 hasConcept C121332964 @default.
- W3153940464 hasConcept C124101348 @default.
- W3153940464 hasConcept C125411270 @default.
- W3153940464 hasConcept C127413603 @default.
- W3153940464 hasConcept C138885662 @default.
- W3153940464 hasConcept C140779682 @default.
- W3153940464 hasConcept C147176958 @default.
- W3153940464 hasConcept C153294291 @default.
- W3153940464 hasConcept C154945302 @default.
- W3153940464 hasConcept C2776401178 @default.
- W3153940464 hasConcept C2777211547 @default.
- W3153940464 hasConcept C2777831296 @default.
- W3153940464 hasConcept C31972630 @default.
- W3153940464 hasConcept C41008148 @default.
- W3153940464 hasConcept C41895202 @default.
- W3153940464 hasConcept C50644808 @default.
- W3153940464 hasConcept C98045186 @default.
- W3153940464 hasConceptScore W3153940464C106131492 @default.
- W3153940464 hasConceptScore W3153940464C111919701 @default.
- W3153940464 hasConceptScore W3153940464C118505674 @default.
- W3153940464 hasConceptScore W3153940464C119857082 @default.
- W3153940464 hasConceptScore W3153940464C121332964 @default.
- W3153940464 hasConceptScore W3153940464C124101348 @default.
- W3153940464 hasConceptScore W3153940464C125411270 @default.
- W3153940464 hasConceptScore W3153940464C127413603 @default.
- W3153940464 hasConceptScore W3153940464C138885662 @default.
- W3153940464 hasConceptScore W3153940464C140779682 @default.
- W3153940464 hasConceptScore W3153940464C147176958 @default.
- W3153940464 hasConceptScore W3153940464C153294291 @default.
- W3153940464 hasConceptScore W3153940464C154945302 @default.
- W3153940464 hasConceptScore W3153940464C2776401178 @default.
- W3153940464 hasConceptScore W3153940464C2777211547 @default.
- W3153940464 hasConceptScore W3153940464C2777831296 @default.
- W3153940464 hasConceptScore W3153940464C31972630 @default.
- W3153940464 hasConceptScore W3153940464C41008148 @default.
- W3153940464 hasConceptScore W3153940464C41895202 @default.
- W3153940464 hasConceptScore W3153940464C50644808 @default.
- W3153940464 hasConceptScore W3153940464C98045186 @default.
- W3153940464 hasLocation W31539404641 @default.
- W3153940464 hasLocation W31539404642 @default.
- W3153940464 hasOpenAccess W3153940464 @default.
- W3153940464 hasPrimaryLocation W31539404641 @default.
- W3153940464 hasRelatedWork W2356875448 @default.
- W3153940464 hasRelatedWork W2358489738 @default.
- W3153940464 hasRelatedWork W2368072106 @default.
- W3153940464 hasRelatedWork W2542782970 @default.
- W3153940464 hasRelatedWork W2547835662 @default.
- W3153940464 hasRelatedWork W2961085424 @default.
- W3153940464 hasRelatedWork W3014617071 @default.
- W3153940464 hasRelatedWork W3163531354 @default.
- W3153940464 hasRelatedWork W4362653130 @default.
- W3153940464 hasRelatedWork W1629725936 @default.