Reasoning Over Virtual Knowledge Bases With Open Predicate Relations

Authors: Haitian Sun, Patrick Verga, Bhuwan Dhingra, Ruslan Salakhutdinov, William W Cohen

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate that OPQL outperforms prior VKB methods on two different KB reasoning tasks and, additionally, can be used as an external memory integrated into a language model (OPQL-LM) leading to improvements on two open-domain question answering tasks.
Researcher Affiliation Collaboration 1Carnegie Mellon University 2Google Research.
Pseudocode No The paper describes methods and equations but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement or link indicating that the source code for the methodology is openly available.
Open Datasets Yes We use Wikipedia passages with hyperlinks as our pretraining. We use the entire Wikipedia as our text corpus for the open-domain QA tasks. Meta QA (Zhang et al., 2018) is a multi-hop QA dataset... Multi-hop Slot Filling (MSF) (Dhingra et al., 2020) presents a large scale multi-hop reasoning dataset... Web Question SP (Web QSP) (Yih et al., 2015) is an open-domain Question Answering dataset... Complex Web Questions (Complex Web Q) (Talmor & Berant, 2018) extends Web Questions SP to multi-hop questions.
Dataset Splits Yes We end up with 10K finetuning data for Meta QA and 19K for MSF. Complex Web Questions (Complex Web Q) (Talmor & Berant, 2018) extends Web Questions SP to multi-hop questions. ... (dev)
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., CPU/GPU models, memory).
Software Dependencies No The paper does not provide specific software dependencies with version numbers needed to reproduce the experiment.
Experiment Setup No The paper provides some high-level finetuning details and mentions that parameters are from Verga et al. (2020) but does not provide specific numerical hyperparameters or system-level training settings within the main text.