Matches in SemOpenAlex for { <https://semopenalex.org/work/W4378501782> ?p ?o ?g. }
Showing items 1 to 83 of
83
with 100 items per page.
- W4378501782 abstract "What can be learned about causality and experimentation from passive data? This question is salient given recent successes of passively-trained language models in interactive domains such as tool use. Passive learning is inherently limited. However, we show that purely passive learning can in fact allow an agent to learn generalizable strategies for determining and using causal structures, as long as the agent can intervene at test time. We formally illustrate that learning a strategy of first experimenting, then seeking goals, can allow generalization from passive learning in principle. We then show empirically that agents trained via imitation on expert data can indeed generalize at test time to infer and use causal links which are never present in the training data; these agents can also generalize experimentation strategies to novel variable sets never observed in training. We then show that strategies for causal intervention and exploitation can be generalized from passive data even in a more complex environment with high-dimensional observations, with the support of natural language explanations. Explanations can even allow passive learners to generalize out-of-distribution from perfectly-confounded training data. Finally, we show that language models, trained only on passive next-word prediction, can generalize causal intervention strategies from a few-shot prompt containing examples of experimentation, together with explanations and reasoning. These results highlight the surprising power of passive learning of active causal strategies, and may help to understand the behaviors and capabilities of language models." @default.
- W4378501782 created "2023-05-27" @default.
- W4378501782 creator A5021342036 @default.
- W4378501782 creator A5030015839 @default.
- W4378501782 creator A5054833182 @default.
- W4378501782 creator A5079236993 @default.
- W4378501782 creator A5082293670 @default.
- W4378501782 date "2023-05-25" @default.
- W4378501782 modified "2023-10-05" @default.
- W4378501782 title "Passive learning of active causal strategies in agents and language models" @default.
- W4378501782 doi "https://doi.org/10.48550/arxiv.2305.16183" @default.
- W4378501782 hasPublicationYear "2023" @default.
- W4378501782 type Work @default.
- W4378501782 citedByCount "0" @default.
- W4378501782 crossrefType "posted-content" @default.
- W4378501782 hasAuthorship W4378501782A5021342036 @default.
- W4378501782 hasAuthorship W4378501782A5030015839 @default.
- W4378501782 hasAuthorship W4378501782A5054833182 @default.
- W4378501782 hasAuthorship W4378501782A5079236993 @default.
- W4378501782 hasAuthorship W4378501782A5082293670 @default.
- W4378501782 hasBestOaLocation W43785017821 @default.
- W4378501782 hasConcept C105795698 @default.
- W4378501782 hasConcept C111472728 @default.
- W4378501782 hasConcept C11671645 @default.
- W4378501782 hasConcept C118552586 @default.
- W4378501782 hasConcept C119857082 @default.
- W4378501782 hasConcept C121332964 @default.
- W4378501782 hasConcept C126388530 @default.
- W4378501782 hasConcept C134306372 @default.
- W4378501782 hasConcept C138885662 @default.
- W4378501782 hasConcept C151730666 @default.
- W4378501782 hasConcept C154945302 @default.
- W4378501782 hasConcept C15744967 @default.
- W4378501782 hasConcept C177148314 @default.
- W4378501782 hasConcept C195324797 @default.
- W4378501782 hasConcept C2777267654 @default.
- W4378501782 hasConcept C2778136018 @default.
- W4378501782 hasConcept C2780665704 @default.
- W4378501782 hasConcept C33923547 @default.
- W4378501782 hasConcept C41008148 @default.
- W4378501782 hasConcept C62520636 @default.
- W4378501782 hasConcept C64357122 @default.
- W4378501782 hasConcept C77805123 @default.
- W4378501782 hasConcept C86803240 @default.
- W4378501782 hasConceptScore W4378501782C105795698 @default.
- W4378501782 hasConceptScore W4378501782C111472728 @default.
- W4378501782 hasConceptScore W4378501782C11671645 @default.
- W4378501782 hasConceptScore W4378501782C118552586 @default.
- W4378501782 hasConceptScore W4378501782C119857082 @default.
- W4378501782 hasConceptScore W4378501782C121332964 @default.
- W4378501782 hasConceptScore W4378501782C126388530 @default.
- W4378501782 hasConceptScore W4378501782C134306372 @default.
- W4378501782 hasConceptScore W4378501782C138885662 @default.
- W4378501782 hasConceptScore W4378501782C151730666 @default.
- W4378501782 hasConceptScore W4378501782C154945302 @default.
- W4378501782 hasConceptScore W4378501782C15744967 @default.
- W4378501782 hasConceptScore W4378501782C177148314 @default.
- W4378501782 hasConceptScore W4378501782C195324797 @default.
- W4378501782 hasConceptScore W4378501782C2777267654 @default.
- W4378501782 hasConceptScore W4378501782C2778136018 @default.
- W4378501782 hasConceptScore W4378501782C2780665704 @default.
- W4378501782 hasConceptScore W4378501782C33923547 @default.
- W4378501782 hasConceptScore W4378501782C41008148 @default.
- W4378501782 hasConceptScore W4378501782C62520636 @default.
- W4378501782 hasConceptScore W4378501782C64357122 @default.
- W4378501782 hasConceptScore W4378501782C77805123 @default.
- W4378501782 hasConceptScore W4378501782C86803240 @default.
- W4378501782 hasLocation W43785017821 @default.
- W4378501782 hasOpenAccess W4378501782 @default.
- W4378501782 hasPrimaryLocation W43785017821 @default.
- W4378501782 hasRelatedWork W1531601525 @default.
- W4378501782 hasRelatedWork W2018580387 @default.
- W4378501782 hasRelatedWork W2574301230 @default.
- W4378501782 hasRelatedWork W2987568073 @default.
- W4378501782 hasRelatedWork W3135588948 @default.
- W4378501782 hasRelatedWork W3170261037 @default.
- W4378501782 hasRelatedWork W3183948672 @default.
- W4378501782 hasRelatedWork W4210420802 @default.
- W4378501782 hasRelatedWork W4238853882 @default.
- W4378501782 hasRelatedWork W4312269093 @default.
- W4378501782 isParatext "false" @default.
- W4378501782 isRetracted "false" @default.
- W4378501782 workType "article" @default.