<i>Pat-in-the-Loop</i>: Declarative Knowledge for Controlling Neural Networks
The dazzling success of neural networks over natural language processing systems is imposing an urgent need to control their behavior with simpler, more direct declarative rules. In this paper, we propose Pat-in-the-Loop as a model to control a specific class of syntax-oriented neural networks by ad...
Main Authors: | , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
MDPI AG
2020-12-01
|
Series: | Future Internet |
Subjects: | |
Online Access: | https://www.mdpi.com/1999-5903/12/12/218 |
id |
doaj-6ace03a84a334f8aa0c25415424f8a6f |
---|---|
record_format |
Article |
spelling |
doaj-6ace03a84a334f8aa0c25415424f8a6f2020-12-03T00:01:20ZengMDPI AGFuture Internet1999-59032020-12-011221821810.3390/fi12120218<i>Pat-in-the-Loop</i>: Declarative Knowledge for Controlling Neural NetworksDario Onorati0Pierfrancesco Tommasino1Leonardo Ranaldi2Francesca Fallucchi3Fabio Massimo Zanzotto4Department of Enterprise Engineering, University of Rome Tor Vergata, 00133 Roma, ItalyDepartment of Enterprise Engineering, University of Rome Tor Vergata, 00133 Roma, ItalyDepartment of Innovation and Information Engineering, Guglielmo Marconi University, 00193 Roma, ItalyDepartment of Innovation and Information Engineering, Guglielmo Marconi University, 00193 Roma, ItalyDepartment of Enterprise Engineering, University of Rome Tor Vergata, 00133 Roma, ItalyThe dazzling success of neural networks over natural language processing systems is imposing an urgent need to control their behavior with simpler, more direct declarative rules. In this paper, we propose Pat-in-the-Loop as a model to control a specific class of syntax-oriented neural networks by adding declarative rules. In Pat-in-the-Loop, <i>distributed tree encoders</i> allow to exploit parse trees in neural networks, <i>heat parse trees</i> visualize activation of parse trees, and parse subtrees are used as declarative rules in the neural network. Hence, Pat-in-the-Loop is a model to include human control in specific natural language processing (NLP)-neural network (NN) systems that exploit syntactic information, which we will generically call Pat. A pilot study on question classification showed that declarative rules representing human knowledge, injected by Pat, can be effectively used in these neural networks to ensure correctness, relevance, and cost-effective.https://www.mdpi.com/1999-5903/12/12/218NLPmachine learningdeep learningAIhuman-in-the-loop |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Dario Onorati Pierfrancesco Tommasino Leonardo Ranaldi Francesca Fallucchi Fabio Massimo Zanzotto |
spellingShingle |
Dario Onorati Pierfrancesco Tommasino Leonardo Ranaldi Francesca Fallucchi Fabio Massimo Zanzotto <i>Pat-in-the-Loop</i>: Declarative Knowledge for Controlling Neural Networks Future Internet NLP machine learning deep learning AI human-in-the-loop |
author_facet |
Dario Onorati Pierfrancesco Tommasino Leonardo Ranaldi Francesca Fallucchi Fabio Massimo Zanzotto |
author_sort |
Dario Onorati |
title |
<i>Pat-in-the-Loop</i>: Declarative Knowledge for Controlling Neural Networks |
title_short |
<i>Pat-in-the-Loop</i>: Declarative Knowledge for Controlling Neural Networks |
title_full |
<i>Pat-in-the-Loop</i>: Declarative Knowledge for Controlling Neural Networks |
title_fullStr |
<i>Pat-in-the-Loop</i>: Declarative Knowledge for Controlling Neural Networks |
title_full_unstemmed |
<i>Pat-in-the-Loop</i>: Declarative Knowledge for Controlling Neural Networks |
title_sort |
<i>pat-in-the-loop</i>: declarative knowledge for controlling neural networks |
publisher |
MDPI AG |
series |
Future Internet |
issn |
1999-5903 |
publishDate |
2020-12-01 |
description |
The dazzling success of neural networks over natural language processing systems is imposing an urgent need to control their behavior with simpler, more direct declarative rules. In this paper, we propose Pat-in-the-Loop as a model to control a specific class of syntax-oriented neural networks by adding declarative rules. In Pat-in-the-Loop, <i>distributed tree encoders</i> allow to exploit parse trees in neural networks, <i>heat parse trees</i> visualize activation of parse trees, and parse subtrees are used as declarative rules in the neural network. Hence, Pat-in-the-Loop is a model to include human control in specific natural language processing (NLP)-neural network (NN) systems that exploit syntactic information, which we will generically call Pat. A pilot study on question classification showed that declarative rules representing human knowledge, injected by Pat, can be effectively used in these neural networks to ensure correctness, relevance, and cost-effective. |
topic |
NLP machine learning deep learning AI human-in-the-loop |
url |
https://www.mdpi.com/1999-5903/12/12/218 |
work_keys_str_mv |
AT darioonorati ipatintheloopideclarativeknowledgeforcontrollingneuralnetworks AT pierfrancescotommasino ipatintheloopideclarativeknowledgeforcontrollingneuralnetworks AT leonardoranaldi ipatintheloopideclarativeknowledgeforcontrollingneuralnetworks AT francescafallucchi ipatintheloopideclarativeknowledgeforcontrollingneuralnetworks AT fabiomassimozanzotto ipatintheloopideclarativeknowledgeforcontrollingneuralnetworks |
_version_ |
1724401738174169088 |