Learning Transferable Features For Open-Domain Question Answering

Zuin, Gianlucca; Chaimowicz, Luiz; Veloso, Adriano

doi:10.1109/ijcnn.2018.8489057

Cited by 12 publications

(4 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…• It is a domain-specific problem, i.e., a prediction model learned from a sub-population (or ICU domain) is likely to fail when tested against data from other population [32]. Feature transferability is thus an appealing way to provide robustness to the prediction models [33], [34].…”

Section: Methodsmentioning

confidence: 99%

Dynamic Prediction of ICU Mortality Risk Using Domain Adaptation

Alves

Laender

Veloso

et al. 2018

2018 IEEE International Conference on Big Data (Big Data)

Self Cite

View full text Add to dashboard Cite

Early recognition of risky trajectories during an Intensive Care Unit (ICU) stay is one of the key steps towards improving patient survival. Learning trajectories from physiological signals continuously measured during an ICU stay requires learning time-series features that are robust and discriminative across diverse patient populations. Patients within different ICU populations (referred here as domains) vary by age, conditions and interventions. Thus, mortality prediction models using patient data from a particular ICU population may perform suboptimally in other populations because the features used to train such models have different distributions across the groups. In this paper, we explore domain adaptation strategies in order to learn mortality prediction models that extract and transfer complex temporal features from multivariate time-series ICU data. Features are extracted in a way that the state of the patient in a certain time depends on the previous state. This enables dynamic predictions and creates a mortality risk space that describes the risk of a patient at a particular time. Experiments based on cross-ICU populations reveals that our model outperforms all considered baselines. Gains in terms of AUC range from 4% to 8% for early predictions when compared with a recent state-of-the-art representative for ICU mortality prediction. In particular, models for the Cardiac ICU population achieve AUC numbers as high as 0.88, showing excellent clinical utility for early mortality prediction. Finally, we present an explanation of factors contributing to the possible ICU outcomes, so that our models can be used to complement clinical reasoning.

show abstract

Section: Methodsmentioning

confidence: 99%

Dynamic Prediction of ICU Mortality Risk Using Domain Adaptation

Alves

Laender

Veloso

et al. 2018

2018 IEEE International Conference on Big Data (Big Data)

Self Cite

View full text Add to dashboard Cite

show abstract

“…To construct the adjacency matrix A, for each node pair (v i , v j ), we applied cosine similarity based on enriched TFIDF features 3 as the value A ij . Previous work has applied cosine similarity for vector space models (García-Pablos et al, 2018;Zuin et al, 2018;Bhatia et al, 2016), so we believe it is a suitable method in our case. This way we were able to generate concept-resource edge values (A cr ) and resourceresource edge values (A r ).…”

Section: Adjacency Matrix Amentioning

confidence: 98%

R-VGAE: Relational-variational Graph Autoencoder for Unsupervised Prerequisite Chain Learning

Li¹,

Fabbri²,

Hingmire³

et al. 2020

Proceedings of the 28th International Conference on Computational Linguistics

View full text Add to dashboard Cite

The task of concept prerequisite chain learning is to automatically determine the existence of prerequisite relationships among concept pairs. In this paper, we frame learning prerequisite relationships among concepts as an unsupervised task with no access to labeled concept pairs during training. We propose a model called the Relational-variational Graph AutoEncoder (R-VGAE) to predict concept relations within a graph consisting of concept and resource nodes. Results show that our unsupervised approach outperforms graph-based semi-supervised methods and other baseline methods by up to 9.77% and 10.47% in terms of prerequisite relation prediction accuracy and F1 score. Our method is notably the first graph-based model that attempts to make use of deep learning representations for the task of unsupervised prerequisite learning. We also expand an existing corpus which totals 1, 717 English Natural Language Processing (NLP)-related lecture slide files and manual concept pair annotations over 322 topics.

show abstract

“…In all our experiments, we constrained the Rashomon sets to include only models subject to the same learning algorithms, such as the hypothesis space of all decision trees with depth up to some value. However, employing different methods enables a final model to capture nuances that are particular to all base methods, such as combining a convolutional neural network and an LSTM to capture both temporal and context-aware patterns [Zuin et al, 2018]. However, our method relies on computing feature importance to compare models.…”

Section: Future Workmentioning

confidence: 99%

Ensemble Learning through Rashomon Sets

Zuin,

Veloso

2024

Anais Do XXXVII Concurso De Teses E Dissertações (CTD 2024)

View full text Add to dashboard Cite

Criar modelos a partir de observações e garantir sua eficácia em novos dados é a essencia do aprendizado de máquina. No entanto, selecionar modelos que generalizem bem para dados futuros continua sendo uma tarefa desafiadora. Neste trabalho, investigamos como os modelos se comportam em conjuntos de dados com funções de geração de dados distintas mas ainda correlacionadas. A motivação é estudar o Efeito Rashomon, que ocorre quando um problema admite a existência de vários modelos distintos com desempenho semelhante. Problemas do mundo real frequentemente exibem múltiplas estruturas locais nos dados, resultando em múltiplos modelos de alto desempenho sujeitos ao Efeito Rashomon. Propomos estratificar durante treino o espaço de soluções em grupos de modelos coerentes ou contrastantes. A partir desses grupos de Rashomon, contruimos um comitê onde cada constituinte cobre uma região distinta do espaço. Validamos nossa abordagem em conjuntos de dados abertos e reais. Nossa abordagem supera o estado-da-arte, melhorando a AUROC em até 0,20+ nos cenários onde a razão de Rashomon é alta.

show abstract

Learning Transferable Features For Open-Domain Question Answering

Cited by 12 publications

References 15 publications

Dynamic Prediction of ICU Mortality Risk Using Domain Adaptation

Dynamic Prediction of ICU Mortality Risk Using Domain Adaptation

R-VGAE: Relational-variational Graph Autoencoder for Unsupervised Prerequisite Chain Learning

Ensemble Learning through Rashomon Sets

Contact Info

Product

Resources

About