Research ArticleSeismology

# Convolutional neural network for earthquake detection and location

See allHide authors and affiliations

Vol. 4, no. 2, e1700578

eLetters is an online forum for ongoing peer review. Submission of eLetters are open to all . Please read our guidelines before submitting your own eLetter.

Compose eLetter

### Plain text

• Plain text
No HTML tags allowed.
• Lines and paragraphs break automatically.
Author Information
Statement of Competing Interests

This question is for testing whether or not you are a human visitor and to prevent automated spam submissions.

## Vertical Tabs

• RE: There may be a typo in equation (1) in the paper "Convolutional neural network for earthquake detection and location"
• Qiwen Zhu, doctoral student, Peking University

Thibaut Perol, et al., wrote an excellent paper "Convolutional neural network for earthquake detection and location". In this paper, they cast earthquake detection as a supervised classification problem and propose the first convolutional neural network for earthquake detection and location (ConvNetQuake).

I learned a lot from their paper. However, I found myself a little bit confused by equation (1) in the paper. After carefully reading the codes the authors used in the paper (https://github.com/tperol/ConvNetQuake) and the documentation of TensorFlow about how the function tensorflow.nn.conv1d (https://tensorflow.google.cn/api_docs/python/tf/nn/conv1d) works, I think the upper bound of the summation over variable c' should be C_(i-1) instead of C_i. For example, when i = 0, the variable c' should change from 1 to C_0 (=3) instead of changing from 1 to C_1 (=32).

Competing Interests: None declared.
• Using not only “Ensemble methods” but also “Ensemble of ensembles methods” play a key role in making more robust model.

Thibaut Perol, et al, write a paper “convolutional neural network for earthquake detection and location” in Science (1). In this article, they try to predict the occurrence and place of earthquakes with a convolutional neural network (1). However, ensemble methods can make more robust prediction model.

Yoshiyasu Takefuji suggests that ensemble methods of machine learning algorithm, including Adaboost, Random forest, and ExtraTree, using scikit-learn (python library) make prediction better (2,3). In addition, ensemble of ensembles methods like Voting-classifier algorithm can also improve prediction than single ensemble algorithm (2,3).

As for ensemble machine learning of convolutional neural networks, I recommend averaging algorithm, especially “Coupled ensemble method” proposed by Anuvabh Dutt et al., (4). It is composed of a few branches of a block like ResNet and DenseNet. In their experiment, it makes better prediction (97.08%) than a single DenseNet or ResNet (the accuracy of each model is approximately 95～96%) on cifar10 image dataset (4). Also, Ensemble of coupled ensembles method, a few branches of coupled ensembles, gets a little higher accuracy (97.32%) than a single coupled ensemble method on the same dataset (4).

To get a model as robust as possible, these two methods should be used, single ensemble approach and ensemble of ensembles way.

References:

1. Thibaut Perol, et al., “convolutional neural network for earthquake detecti...