Data-driven and active learning of variance-based sensitivity indices with Bayesian probabilistic integration

The main objective of this paper is assessing the usefulness of parameter sensitivity information from conceptual hydrological models for data-driven models, an approach which might allow us to take advantage of the strengths of both data-based and process-based models. This study uses the parameter sensitivity of three widely used conceptual hydrological models (GR4J, Hymod and SAC-SMA) and combines them with M5 model trees. The study was carried out for three case studies dealing with different problems to which model trees are applied: one using model trees as error correctors and two case studies in which model trees were used as rainfall–runoff models and which differ in how the sensitivity information is used. The results show that sensitivity time series can improve the predictions of M5 model trees, especially when they do not include the time series of previous discharge as predictor variables. The use of parameter sensitivity information for clustering the time series resulted in model trees that had a structure consistent with the hydrological processes that were taking place in the considered cluster, indicating that the use of sensitivity indices could be a viable way of introducing hydrological knowledge into data-based models.

Download Full-text

CAiRE: An End-to-End Empathetic Chatbot

Proceedings of the AAAI Conference on Artificial Intelligence ◽

10.1609/aaai.v34i09.7098 ◽

2020 ◽

Vol 34 (09) ◽

pp. 13622-13623

Author(s):

Zhaojiang Lin ◽

Peng Xu ◽

Genta Indra Winata ◽

Farhad Bin Siddique ◽

Zihan Liu ◽

...

Keyword(s):

User Interface ◽

Active Learning ◽

Transfer Learning ◽

User Feedback ◽

Data Driven ◽

Dialogue Systems ◽

Generation Task ◽

Web Based ◽

End To End ◽

Negative Training

We present CAiRE, an end-to-end generative empathetic chatbot designed to recognize user emotions and respond in an empathetic manner. Our system adapts the Generative Pre-trained Transformer (GPT) to empathetic response generation task via transfer learning. CAiRE is built primarily to focus on empathy integration in fully data-driven generative dialogue systems. We create a web-based user interface which allows multiple users to asynchronously chat with CAiRE. CAiRE also collects user feedback and continues to improve its response quality by discarding undesirable generations via active learning and negative training.

Download Full-text

Data-driven experimental design and model development using Gaussian process with active learning

Cognitive Psychology ◽

10.1016/j.cogpsych.2020.101360 ◽

2021 ◽

Vol 125 ◽

pp. 101360

Author(s):

Jorge Chang ◽

Jiseob Kim ◽

Byoung-Tak Zhang ◽

Mark A. Pitt ◽

Jay I. Myung

Keyword(s):

Experimental Design ◽

Active Learning ◽

Gaussian Process ◽

Model Development ◽

Data Driven

Download Full-text

Data-driven Verification using Efficient Active Learning

10.2514/6.2022-2086 ◽

2022 ◽

Author(s):

Venkata Vaishnav Tadiparthi ◽

Raktim Bhattacharya

Keyword(s):

Active Learning ◽

Data Driven

Download Full-text

Using Pre-trained Language Model to Enhance Active Learning for Sentence Matching

ACM Transactions on Asian and Low-Resource Language Information Processing ◽

10.1145/3480937 ◽

2022 ◽

Vol 21 (2) ◽

pp. 1-19

Author(s):

Guirong Bai ◽

Shizhu He ◽

Kang Liu ◽

Jun Zhao

Keyword(s):

Active Learning ◽

Language Model ◽

Experimental Results ◽

Language Models ◽

Data Driven ◽

Learning Approach ◽

Sentence Matching ◽

Learning Language

Active learning is an effective method to substantially alleviate the problem of expensive annotation cost for data-driven models. Recently, pre-trained language models have been demonstrated to be powerful for learning language representations. In this article, we demonstrate that the pre-trained language model can also utilize its learned textual characteristics to enrich criteria of active learning. Specifically, we provide extra textual criteria with the pre-trained language model to measure instances, including noise, coverage, and diversity. With these extra textual criteria, we can select more efficient instances for annotation and obtain better results. We conduct experiments on both English and Chinese sentence matching datasets. The experimental results show that the proposed active learning approach can be enhanced by the pre-trained language model and obtain better performance.

Download Full-text