Filters








29,167 Hits in 6.4 sec

Logic-Guided Data Augmentation and Regularization for Consistent Question Answering [article]

Akari Asai, Hannaneh Hajishirzi
2020 arXiv   pre-print
Our method leverages logical and linguistic knowledge to augment labeled training data and then uses a consistency-based regularizer to train the model.  ...  This paper addresses the problem of improving the accuracy and consistency of responses to comparison questions by integrating logic rules and neural models.  ...  We thank Antoine Bosselut, Tim Dettmers, Rik Koncel-Kedziorski, Sewon Min, Keisuke Sakaguchi, David Wadden, Yizhong Wang, the members of UW NLP group and AI2, and the anonymous reviewers for their insightful  ... 
arXiv:2004.10157v2 fatcat:5hbhcyawirdbbhqxic7qlqb6ou

Logic-Guided Data Augmentation and Regularization for Consistent Question Answering

Akari Asai, Hannaneh Hajishirzi
2020 Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics   unpublished
Our method leverages logical and linguistic knowledge to augment labeled training data and then uses a consistency-based regularizer to train the model.  ...  This paper addresses the problem of improving the accuracy and consistency of responses to comparison questions by integrating logic rules and neural models.  ...  We thank Antoine Bosselut, Tim Dettmers, Rik Koncel-Kedziorski, Sewon Min, Keisuke Sakaguchi, David Wadden, Yizhong Wang, the members of UW NLP group and AI2, and the anonymous reviewers for their insightful  ... 
doi:10.18653/v1/2020.acl-main.499 fatcat:6zdjzhxuk5e6hlpnpzdhy63u74

Logically Consistent Loss for Visual Question Answering [article]

Anh-Cat Le-Ngo, Truyen Tran, Santu Rana, Sunil Gupta, Svetha Venkatesh
2020 arXiv   pre-print
Given an image, a back-ground knowledge, and a set of questions about an object, human learners answer the questions very consistently regardless of question forms and semantic tasks.  ...  To demonstrate usefulness of this proposal, we train and evaluate MAC-net based VQA machines with and without the proposed logically consistent loss and the proposed data organization.  ...  [17] use data-augmentation approach to enforce consistency between pairs of questions and answers, Selvaraju et al.  ... 
arXiv:2011.10094v1 fatcat:mjiflqckvbd33plsq6ldpv4qty

Augmenting Neural Networks with First-order Logic

Tao Li, Vivek Srikumar
2019 Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics  
In this paper, we present a novel framework for introducing declarative knowledge to neural network architectures in order to guide training and prediction.  ...  Our experiments show that knowledge-augmented networks can strongly improve over baselines, especially in low-data regimes.  ...  Acknowledgements We thank members of the NLP group at the University of Utah for their valuable insights and suggestions; and reviewers for pointers to related works, corrections, and helpful comments.  ... 
doi:10.18653/v1/p19-1028 dblp:conf/acl/LiS19 fatcat:fd267yehgnggphq7jtznfzi2la

GraPPa: Grammar-Augmented Pre-Training for Table Semantic Parsing [article]

Tao Yu and Chien-Sheng Wu and Xi Victoria Lin and Bailin Wang and Yi Chern Tan and Xinyi Yang and Dragomir Radev and Richard Socher and Caiming Xiong
2021 arXiv   pre-print
We pre-train our model on the synthetic data using a novel text-schema linking objective that predicts the syntactic role of a table field in the SQL for each question-SQL pair.  ...  We present GraPPa, an effective pre-training approach for table semantic parsing that learns a compositional inductive bias in the joint representations of textual and tabular data.  ...  However, data augmentation becomes more complex and less beneficial if we want to apply it to generate data for a random domain.  ... 
arXiv:2009.13845v2 fatcat:dzy5bu2cbna2lg7m6zun4lavhm

Augmenting Neural Networks with First-order Logic [article]

Tao Li, Vivek Srikumar
2020 arXiv   pre-print
In this paper, we present a novel framework for introducing declarative knowledge to neural network architectures in order to guide training and prediction.  ...  Our experiments show that knowledge-augmented networks can strongly improve over baselines, especially in low-data regimes.  ...  Acknowledgements We thank members of the NLP group at the University of Utah for their valuable insights and suggestions; and reviewers for pointers to related works, corrections, and helpful comments.  ... 
arXiv:1906.06298v3 fatcat:o4i7fbgvmrc5hlmi5zhoawfg7q

Learning from Lexical Perturbations for Consistent Visual Question Answering [article]

Spencer Whitehead, Hui Wu, Yi Ren Fung, Heng Ji, Rogerio Feris, Kate Saenko
2020 arXiv   pre-print
regularization tool for VQA models.  ...  Existing Visual Question Answering (VQA) models are often fragile and sensitive to input variations.  ...  To summarize, our main contributions are: • A novel VQA consistency regularization method that augments questions and enforces similar answers and reasoning steps for the original and augmented questions  ... 
arXiv:2011.13406v2 fatcat:poyfejkn4nbx3h3t6pkzdlbzly

Iterative Search for Weakly Supervised Semantic Parsing

Pradeep Dasigi, Matt Gardner, Shikhar Murty, Luke Zettlemoyer, Eduard Hovy
2019 Proceedings of the 2019 Conference of the North  
We propose a novel iterative training algorithm that alternates between searching for consistent logical forms and maximizing the marginal likelihood of the retrieved ones.  ...  that coincidentally evaluate to the correct answer.  ...  Acknowledgments We would like to thank Jonathan Berant and Noah Smith for comments on earlier drafts and Chen Liang for helping us with implementation details of MAPO.  ... 
doi:10.18653/v1/n19-1273 dblp:conf/naacl/Dasigi0MZH19 fatcat:ow7ygq7sqbdwjdmcw2bbwgbdqa

Compositional Semantic Parsing on Semi-Structured Tables

Panupong Pasupat, Percy Liang
2015 Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)  
Two important aspects of semantic parsing for question answering are the breadth of the knowledge source and the depth of logical compositionality.  ...  We propose a logical-form driven parsing algorithm guided by strong typing constraints and show that it obtains significant improvements over natural baselines.  ...  Additionally, code, data, and experiments for this paper are available on the CodaLab platform at https://www.codalab.org/worksheets/ 0xf26cd79d4d734287868923ad1067cf4c/.  ... 
doi:10.3115/v1/p15-1142 dblp:conf/acl/PasupatL15 fatcat:d5xqyxrcmrfshoxka7uaed4yku

Compositional Semantic Parsing on Semi-Structured Tables [article]

Panupong Pasupat, Percy Liang
2015 arXiv   pre-print
Two important aspects of semantic parsing for question answering are the breadth of the knowledge source and the depth of logical compositionality.  ...  We propose a logical-form driven parsing algorithm guided by strong typing constraints and show that it obtains significant improvements over natural baselines.  ...  Additionally, code, data, and experiments for this paper are available on the CodaLab platform at https://www.codalab.org/worksheets/ 0xf26cd79d4d734287868923ad1067cf4c/.  ... 
arXiv:1508.00305v1 fatcat:iyqdr3ikjrfmli5wqzxomqznqm

A Closer Look at the Robustness of Vision-and-Language Pre-trained Models [article]

Linjie Li, Zhe Gan, Jingjing Liu
2021 arXiv   pre-print
Visual Content Manipulation; and (iv) Answer Distribution Shift.  ...  Differing from previous studies focused on one specific type of robustness, Mango is task-agnostic, and enables universal performance lift for pre-trained models over diverse tasks designed to evaluate  ...  It consists of two datasets: VQA-LOL Compose (logical combinations of multiple closed binary questions about the same image in VQA v2) and VQA-LOL Supplement (logical combinations of additional questions  ... 
arXiv:2012.08673v2 fatcat:orl3dt3r3fg3xjac2rt4xwqxxu

Leveraging Declarative Knowledge in Text and First-Order Logic for Fine-Grained Propaganda Detection [article]

Ruize Wang, Duyu Tang, Nan Duan, Wanjun Zhong, Zhongyu Wei, Xuanjing Huang, Daxin Jiang, Ming Zhou
2020 arXiv   pre-print
The former refers to the logical consistency between coarse- and fine-grained predictions, which is used to regularize the training process with propositional Boolean expressions.  ...  Specifically, we leverage the declarative knowledge expressed in both first-order logic and natural language.  ...  Acknowledgments This work is partically supported by National Natural Science Foundation of China (No. 71991471), Science and Technology Commission of Shanghai Municipality Grant (No.20dz1200600, No.18DZ1201000  ... 
arXiv:2004.14201v2 fatcat:vfscs2s2vzb35o3gwlkqq6kwfu

Neuro-Symbolic Entropy Regularization [article]

Kareem Ahmed, Eric Wang, Kai-Wei Chang, Guy Van den Broeck
2022 arXiv   pre-print
Such a large output space makes learning hard and requires vast amounts of labeled data. Different approaches leverage alternate sources of supervision.  ...  and more likely to be valid.  ...  Neuro-symbolic entropy-regularization guides the network to valid and confident predictions (d). unlabeled points, thereby supplementing scarce labeled data with abundant unlabeled data.  ... 
arXiv:2201.11250v1 fatcat:6li4zcdrknaxzc3whcnlkp27qm

Neural Programmer: Inducing Latent Programs with Gradient Descent [article]

Arvind Neelakantan, Quoc V. Le, Ilya Sutskever
2016 arXiv   pre-print
However, this success has not been translated to applications like question answering that may involve complex arithmetic and logic reasoning.  ...  In this work, we propose Neural Programmer, an end-to-end differentiable neural network augmented with a small set of basic arithmetic and logic operations.  ...  Acknowledgements We sincerely thank Greg Corrado, Andrew Dai, Jeff Dean, Shixiang Gu, Andrew McCallum, and Luke Vilnis for their suggestions and the Google Brain team for the support.  ... 
arXiv:1511.04834v3 fatcat:h4rbj7uhvrc5bi2z7qyrgv5yfm

On Incorporating Semantic Prior Knowledge in Deep Learning Through Embedding-Space Constraints [article]

Damien Teney, Ehsan Abbasnejad, Anton van den Hengel
2019 arXiv   pre-print
We illustrate the method on the task of visual question answering to exploit various auxiliary annotations, including relations of equivalence and of logical entailment between questions.  ...  Existing methods to use these annotations, including auxiliary losses and data augmentation, cannot guarantee the strict inclusion of these relations into the model since they require a careful balancing  ...  They used it for data augmentation while ensuring that all generated versions lead to the same answer, i.e. enforcing cycle consistency.  ... 
arXiv:1909.13471v2 fatcat:ccmpn7grkzakfc3nb2tdkzksma
« Previous Showing results 1 — 15 out of 29,167 results