Causal Discovery with Attention-Based Convolutional Neural Networks

Meike Nauta (Corresponding Author), Doina Bucur, Christin Seifert

Research output: Contribution to journalArticleAcademicpeer-review

Abstract

Having insight into the causal associations in a complex system facilitates decision making, e.g., for medical treatments, urban infrastructure improvements or financial investments. The amount of observational data grows, which enables the discovery of causal relationships between variables from observation of their behaviour in time. Existing methods for causal discovery from time series data do not yet exploit the representational power of deep learning. We therefore present the Temporal Causal Discovery Framework (TCDF), a deep learning framework that learns a causal graph structure by discovering causal relationships in observational time series data. TCDF uses attention-based convolutional neural networks combined with a causal validation step. By interpreting the internal parameters of the convolutional networks, TCDF can also discover the time delay between a cause and the occurrence of its effect. Our framework learns temporal causal graphs, which can include confounders and instantaneous effects. Experiments on financial and neuroscientific benchmarks show state-of-the-art performance of TCDF on discovering causal relationships in continuous time series data. Furthermore, we show that TCDF can circumstantially discover the presence of hidden confounders. Our broadly applicable framework can be used to gain novel insights into the causal dependencies in a complex system, which is important for reliable predictions, knowledge discovery and data-driven decision making.
LanguageEnglish
Pages312-340
Number of pages28
JournalMachine Learning and Knowledge Extraction
Volume1
Issue number1
DOIs
Publication statusPublished - 7 Jan 2019

Fingerprint

Time series
Neural networks
Large scale systems
Decision making
Data mining
Time delay
Experiments
Deep learning

Keywords

  • Convolutional Neural Network
  • Time series
  • Causal Discovery
  • Attention
  • Machine Learning

Cite this

@article{94aa7e4899454f4e9e5262243d15a9f8,
title = "Causal Discovery with Attention-Based Convolutional Neural Networks",
abstract = "Having insight into the causal associations in a complex system facilitates decision making, e.g., for medical treatments, urban infrastructure improvements or financial investments. The amount of observational data grows, which enables the discovery of causal relationships between variables from observation of their behaviour in time. Existing methods for causal discovery from time series data do not yet exploit the representational power of deep learning. We therefore present the Temporal Causal Discovery Framework (TCDF), a deep learning framework that learns a causal graph structure by discovering causal relationships in observational time series data. TCDF uses attention-based convolutional neural networks combined with a causal validation step. By interpreting the internal parameters of the convolutional networks, TCDF can also discover the time delay between a cause and the occurrence of its effect. Our framework learns temporal causal graphs, which can include confounders and instantaneous effects. Experiments on financial and neuroscientific benchmarks show state-of-the-art performance of TCDF on discovering causal relationships in continuous time series data. Furthermore, we show that TCDF can circumstantially discover the presence of hidden confounders. Our broadly applicable framework can be used to gain novel insights into the causal dependencies in a complex system, which is important for reliable predictions, knowledge discovery and data-driven decision making.",
keywords = "Convolutional Neural Network, Time series, Causal Discovery, Attention, Machine Learning",
author = "Meike Nauta and Doina Bucur and Christin Seifert",
year = "2019",
month = "1",
day = "7",
doi = "10.3390/make1010019",
language = "English",
volume = "1",
pages = "312--340",
journal = "Machine Learning and Knowledge Extraction",
issn = "2504-4990",
publisher = "Multidisciplinary Digital Publishing Institute",
number = "1",

}

Causal Discovery with Attention-Based Convolutional Neural Networks. / Nauta, Meike (Corresponding Author); Bucur, Doina ; Seifert, Christin .

In: Machine Learning and Knowledge Extraction, Vol. 1, No. 1, 07.01.2019, p. 312-340.

Research output: Contribution to journalArticleAcademicpeer-review

TY - JOUR

T1 - Causal Discovery with Attention-Based Convolutional Neural Networks

AU - Nauta, Meike

AU - Bucur, Doina

AU - Seifert, Christin

PY - 2019/1/7

Y1 - 2019/1/7

N2 - Having insight into the causal associations in a complex system facilitates decision making, e.g., for medical treatments, urban infrastructure improvements or financial investments. The amount of observational data grows, which enables the discovery of causal relationships between variables from observation of their behaviour in time. Existing methods for causal discovery from time series data do not yet exploit the representational power of deep learning. We therefore present the Temporal Causal Discovery Framework (TCDF), a deep learning framework that learns a causal graph structure by discovering causal relationships in observational time series data. TCDF uses attention-based convolutional neural networks combined with a causal validation step. By interpreting the internal parameters of the convolutional networks, TCDF can also discover the time delay between a cause and the occurrence of its effect. Our framework learns temporal causal graphs, which can include confounders and instantaneous effects. Experiments on financial and neuroscientific benchmarks show state-of-the-art performance of TCDF on discovering causal relationships in continuous time series data. Furthermore, we show that TCDF can circumstantially discover the presence of hidden confounders. Our broadly applicable framework can be used to gain novel insights into the causal dependencies in a complex system, which is important for reliable predictions, knowledge discovery and data-driven decision making.

AB - Having insight into the causal associations in a complex system facilitates decision making, e.g., for medical treatments, urban infrastructure improvements or financial investments. The amount of observational data grows, which enables the discovery of causal relationships between variables from observation of their behaviour in time. Existing methods for causal discovery from time series data do not yet exploit the representational power of deep learning. We therefore present the Temporal Causal Discovery Framework (TCDF), a deep learning framework that learns a causal graph structure by discovering causal relationships in observational time series data. TCDF uses attention-based convolutional neural networks combined with a causal validation step. By interpreting the internal parameters of the convolutional networks, TCDF can also discover the time delay between a cause and the occurrence of its effect. Our framework learns temporal causal graphs, which can include confounders and instantaneous effects. Experiments on financial and neuroscientific benchmarks show state-of-the-art performance of TCDF on discovering causal relationships in continuous time series data. Furthermore, we show that TCDF can circumstantially discover the presence of hidden confounders. Our broadly applicable framework can be used to gain novel insights into the causal dependencies in a complex system, which is important for reliable predictions, knowledge discovery and data-driven decision making.

KW - Convolutional Neural Network

KW - Time series

KW - Causal Discovery

KW - Attention

KW - Machine Learning

U2 - 10.3390/make1010019

DO - 10.3390/make1010019

M3 - Article

VL - 1

SP - 312

EP - 340

JO - Machine Learning and Knowledge Extraction

T2 - Machine Learning and Knowledge Extraction

JF - Machine Learning and Knowledge Extraction

SN - 2504-4990

IS - 1

ER -