Matches in SemOpenAlex for { <https://semopenalex.org/work/W4387687923> ?p ?o ?g. }
Showing items 1 to 85 of
85
with 100 items per page.
- W4387687923 abstract "When applied to question answering and other text generation tasks, language models (LMs) may be queried generatively (by sampling answers from their output distribution) or discriminatively (by using them to score or rank a set of candidate outputs). These procedures sometimes yield very different predictions. How do we reconcile mutually incompatible scoring procedures to obtain coherent LM predictions? We introduce a new, a training-free, game-theoretic procedure for language model decoding. Our approach casts language model decoding as a regularized imperfect-information sequential signaling game - which we term the CONSENSUS GAME - in which a GENERATOR seeks to communicate an abstract correctness parameter using natural language sentences to a DISCRIMINATOR. We develop computational procedures for finding approximate equilibria of this game, resulting in a decoding algorithm we call EQUILIBRIUM-RANKING. Applied to a large number of tasks (including reading comprehension, commonsense reasoning, mathematical problem-solving, and dialog), EQUILIBRIUM-RANKING consistently, and sometimes substantially, improves performance over existing LM decoding procedures - on multiple benchmarks, we observe that applying EQUILIBRIUM-RANKING to LLaMA-7B outperforms the much larger LLaMA-65B and PaLM-540B models. These results highlight the promise of game-theoretic tools for addressing fundamental challenges of truthfulness and consistency in LMs." @default.
- W4387687923 created "2023-10-17" @default.
- W4387687923 creator A5016916513 @default.
- W4387687923 creator A5070571735 @default.
- W4387687923 creator A5070829652 @default.
- W4387687923 creator A5073742611 @default.
- W4387687923 date "2023-10-13" @default.
- W4387687923 modified "2023-10-18" @default.
- W4387687923 title "The Consensus Game: Language Model Generation via Equilibrium Search" @default.
- W4387687923 doi "https://doi.org/10.48550/arxiv.2310.09139" @default.
- W4387687923 hasPublicationYear "2023" @default.
- W4387687923 type Work @default.
- W4387687923 citedByCount "0" @default.
- W4387687923 crossrefType "posted-content" @default.
- W4387687923 hasAuthorship W4387687923A5016916513 @default.
- W4387687923 hasAuthorship W4387687923A5070571735 @default.
- W4387687923 hasAuthorship W4387687923A5070829652 @default.
- W4387687923 hasAuthorship W4387687923A5073742611 @default.
- W4387687923 hasBestOaLocation W43876879231 @default.
- W4387687923 hasConcept C11413529 @default.
- W4387687923 hasConcept C114614502 @default.
- W4387687923 hasConcept C119857082 @default.
- W4387687923 hasConcept C121332964 @default.
- W4387687923 hasConcept C137293760 @default.
- W4387687923 hasConcept C154945302 @default.
- W4387687923 hasConcept C163258240 @default.
- W4387687923 hasConcept C164226766 @default.
- W4387687923 hasConcept C177264268 @default.
- W4387687923 hasConcept C189430467 @default.
- W4387687923 hasConcept C195324797 @default.
- W4387687923 hasConcept C199360897 @default.
- W4387687923 hasConcept C204321447 @default.
- W4387687923 hasConcept C2776436953 @default.
- W4387687923 hasConcept C2779439875 @default.
- W4387687923 hasConcept C2779803651 @default.
- W4387687923 hasConcept C2780992000 @default.
- W4387687923 hasConcept C33923547 @default.
- W4387687923 hasConcept C41008148 @default.
- W4387687923 hasConcept C55439883 @default.
- W4387687923 hasConcept C57273362 @default.
- W4387687923 hasConcept C62520636 @default.
- W4387687923 hasConcept C76155785 @default.
- W4387687923 hasConcept C80444323 @default.
- W4387687923 hasConcept C94915269 @default.
- W4387687923 hasConceptScore W4387687923C11413529 @default.
- W4387687923 hasConceptScore W4387687923C114614502 @default.
- W4387687923 hasConceptScore W4387687923C119857082 @default.
- W4387687923 hasConceptScore W4387687923C121332964 @default.
- W4387687923 hasConceptScore W4387687923C137293760 @default.
- W4387687923 hasConceptScore W4387687923C154945302 @default.
- W4387687923 hasConceptScore W4387687923C163258240 @default.
- W4387687923 hasConceptScore W4387687923C164226766 @default.
- W4387687923 hasConceptScore W4387687923C177264268 @default.
- W4387687923 hasConceptScore W4387687923C189430467 @default.
- W4387687923 hasConceptScore W4387687923C195324797 @default.
- W4387687923 hasConceptScore W4387687923C199360897 @default.
- W4387687923 hasConceptScore W4387687923C204321447 @default.
- W4387687923 hasConceptScore W4387687923C2776436953 @default.
- W4387687923 hasConceptScore W4387687923C2779439875 @default.
- W4387687923 hasConceptScore W4387687923C2779803651 @default.
- W4387687923 hasConceptScore W4387687923C2780992000 @default.
- W4387687923 hasConceptScore W4387687923C33923547 @default.
- W4387687923 hasConceptScore W4387687923C41008148 @default.
- W4387687923 hasConceptScore W4387687923C55439883 @default.
- W4387687923 hasConceptScore W4387687923C57273362 @default.
- W4387687923 hasConceptScore W4387687923C62520636 @default.
- W4387687923 hasConceptScore W4387687923C76155785 @default.
- W4387687923 hasConceptScore W4387687923C80444323 @default.
- W4387687923 hasConceptScore W4387687923C94915269 @default.
- W4387687923 hasLocation W43876879231 @default.
- W4387687923 hasOpenAccess W4387687923 @default.
- W4387687923 hasPrimaryLocation W43876879231 @default.
- W4387687923 hasRelatedWork W2049117375 @default.
- W4387687923 hasRelatedWork W2542958340 @default.
- W4387687923 hasRelatedWork W2900126711 @default.
- W4387687923 hasRelatedWork W2967994095 @default.
- W4387687923 hasRelatedWork W3015724364 @default.
- W4387687923 hasRelatedWork W3202115945 @default.
- W4387687923 hasRelatedWork W4225162083 @default.
- W4387687923 hasRelatedWork W4285240985 @default.
- W4387687923 hasRelatedWork W4286930972 @default.
- W4387687923 hasRelatedWork W4288263119 @default.
- W4387687923 isParatext "false" @default.
- W4387687923 isRetracted "false" @default.
- W4387687923 workType "article" @default.