FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3118668786> ?p ?o ?g. }

• W3118668786 abstract "Conducting text retrieval in a dense representation space has many intriguing advantages. Yet the end-to-end learned dense retrieval (DR) often underperforms word-based sparse retrieval. In this paper, we first theoretically show the learning bottleneck of dense retrieval is due to the domination of uninformative negatives sampled locally in batch, which yield diminishing gradient norms, large stochastic gradient variances, and slow learning convergence. We then propose Approximate nearest neighbor Negative Contrastive Learning (ANCE), a learning mechanism that selects hard training negatives globally from the entire corpus, using an asynchronously updated ANN index. Our experiments demonstrate the effectiveness of ANCE on web search, question answering, and in a commercial search environment, showing ANCE dot-product retrieval nearly matches the accuracy of BERT-based cascade IR pipeline, while being 100x more efficient. We also empirically validate our theory that negative sampling with ANCE better approximates the oracle gradient-norm based importance sampling, thus improves the convergence of stochastic training." @default.
• W3118668786 created "2021-01-18" @default.
• W3118668786 creator A5013973657 @default.
• W3118668786 creator A5026109848 @default.
• W3118668786 creator A5030115328 @default.
• W3118668786 creator A5033316313 @default.
• W3118668786 creator A5054440828 @default.
• W3118668786 creator A5068440510 @default.
• W3118668786 creator A5069354508 @default.
• W3118668786 creator A5086696969 @default.
• W3118668786 date "2021-05-03" @default.
• W3118668786 modified "2023-09-23" @default.
• W3118668786 title "Approximate Nearest Neighbor Negative Contrastive Learning for Dense Text Retrieval" @default.
• W3118668786 cites W1973435495 @default.
• W3118668786 cites W2120308175 @default.
• W3118668786 cites W2152790380 @default.
• W3118668786 cites W2177410802 @default.
• W3118668786 cites W2187089797 @default.
• W3118668786 cites W2536015822 @default.
• W3118668786 cites W2648699835 @default.
• W3118668786 cites W2725049817 @default.
• W3118668786 cites W2794302998 @default.
• W3118668786 cites W2798991696 @default.
• W3118668786 cites W2842511635 @default.
• W3118668786 cites W2892283076 @default.
• W3118668786 cites W2909544278 @default.
• W3118668786 cites W2912924812 @default.
• W3118668786 cites W2937036051 @default.
• W3118668786 cites W2951434086 @default.
• W3118668786 cites W2951534261 @default.
• W3118668786 cites W2962985038 @default.
• W3118668786 cites W2963310665 @default.
• W3118668786 cites W2963339397 @default.
• W3118668786 cites W2963499204 @default.
• W3118668786 cites W2963748441 @default.
• W3118668786 cites W2964060837 @default.
• W3118668786 cites W2964110616 @default.
• W3118668786 cites W2965373594 @default.
• W3118668786 cites W2982096936 @default.
• W3118668786 cites W2987283559 @default.
• W3118668786 cites W2988841832 @default.
• W3118668786 cites W2995289474 @default.
• W3118668786 cites W2996064239 @default.
• W3118668786 cites W2998702515 @default.
• W3118668786 cites W3003186568 @default.
• W3118668786 cites W3005680577 @default.
• W3118668786 cites W3007672467 @default.
• W3118668786 cites W3009561768 @default.
• W3118668786 cites W3013936901 @default.
• W3118668786 cites W3021282678 @default.
• W3118668786 cites W3022373106 @default.
• W3118668786 cites W3034696692 @default.
• W3118668786 cites W3098468692 @default.
• W3118668786 cites W3099700870 @default.
• W3118668786 cites W3102378333 @default.
• W3118668786 cites W3102659883 @default.
• W3118668786 cites W3105107530 @default.
• W3118668786 cites W3105721709 @default.
• W3118668786 cites W3105817677 @default.
• W3118668786 cites W3120490999 @default.
• W3118668786 cites W3125238517 @default.
• W3118668786 cites W3178067142 @default.
• W3118668786 hasPublicationYear "2021" @default.
• W3118668786 type Work @default.
• W3118668786 sameAs 3118668786 @default.
• W3118668786 citedByCount "59" @default.
• W3118668786 countsByYear W31186687862020 @default.
• W3118668786 countsByYear W31186687862021 @default.
• W3118668786 countsByYear W31186687862022 @default.
• W3118668786 crossrefType "proceedings-article" @default.
• W3118668786 hasAuthorship W3118668786A5013973657 @default.
• W3118668786 hasAuthorship W3118668786A5026109848 @default.
• W3118668786 hasAuthorship W3118668786A5030115328 @default.
• W3118668786 hasAuthorship W3118668786A5033316313 @default.
• W3118668786 hasAuthorship W3118668786A5054440828 @default.
• W3118668786 hasAuthorship W3118668786A5068440510 @default.
• W3118668786 hasAuthorship W3118668786A5069354508 @default.
• W3118668786 hasAuthorship W3118668786A5086696969 @default.
• W3118668786 hasConcept C113238511 @default.
• W3118668786 hasConcept C115903868 @default.
• W3118668786 hasConcept C119857082 @default.
• W3118668786 hasConcept C149635348 @default.
• W3118668786 hasConcept C153180895 @default.
• W3118668786 hasConcept C154945302 @default.
• W3118668786 hasConcept C161986146 @default.
• W3118668786 hasConcept C199360897 @default.
• W3118668786 hasConcept C2780513914 @default.
• W3118668786 hasConcept C41008148 @default.
• W3118668786 hasConcept C43521106 @default.
• W3118668786 hasConcept C55166926 @default.
• W3118668786 hasConceptScore W3118668786C113238511 @default.
• W3118668786 hasConceptScore W3118668786C115903868 @default.
• W3118668786 hasConceptScore W3118668786C119857082 @default.
• W3118668786 hasConceptScore W3118668786C149635348 @default.
• W3118668786 hasConceptScore W3118668786C153180895 @default.
• W3118668786 hasConceptScore W3118668786C154945302 @default.
• W3118668786 hasConceptScore W3118668786C161986146 @default.
• W3118668786 hasConceptScore W3118668786C199360897 @default.
• W3118668786 hasConceptScore W3118668786C2780513914 @default.
• W3118668786 hasConceptScore W3118668786C41008148 @default.

• next