Figure 4 - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Hand position for the left-hand key set, with the keyboard at an angle of 10-15 • to minimise finger reach strain.
Source publication
The machine-learning research community has focused greatly on bias in
algorithms and have identified different manifestations of it. Bias in
training samples is recognised as a potential source of prejudice
in machine learning. It can be introduced by the human experts who define
the training sets. As machine-learning techniques are being applied...
Context in source publication
Citations
... The network will ideally classify the most dominant auroral feature. However, an evaluation of the performance on ambiguous images will itself be subject to biases and subjective interpretations, as described by McKay and Kvammen (2020). It is irrational to expect a DNN to classify ambiguous images correctly if not even auroral experts can agree on what the correct label is. ...
... The DNNs generally outperformed the KNN and SVM techniques. However, progress in this field of study is constrained by biases and subjective interpretations (McKay & Kvammen, 2020). It is irrational to expect the DNNs to classify an auroral image correctly if auroral researchers cannot agree on what the correct aurora label is. ...
Results from a study of automatic aurora classification using machine learning techniques are presented. The aurora is the manifestation of physical phenomena in the ionosphere‐magnetosphere environment. Automatic classification of millions of auroral images from the Arctic and Antarctic is therefore an attractive tool for developing auroral statistics and for supporting scientists to study auroral images in an objective, organized, and repeatable manner. Although previous studies have presented tools for detecting aurora, there has been a lack of tools for classifying aurora into subclasses with a high precision (>90%). This work considers seven auroral subclasses: breakup, colored, arcs, discrete, patchy, edge, and faint. Six different deep neural network architectures have been tested along with the well‐known classification algorithms: k‐nearest neighbor (KNN) and a support vector machine (SVM). A set of clean nighttime color auroral images, without clearly ambiguous auroral forms, moonlight, twilight, clouds, and so forth, were used for training and testing the classifiers. The deep neural networks generally outperformed the KNN and SVM methods, and the ResNet‐50 architecture achieved the highest performance with an average classification precision of 92%.
We develop an open source algorithm to apply Transfer learning to Aurora image classification and Magnetic disturbance Evaluation (TAME). For this purpose, we evaluate the performance of 80 pretrained neural networks using the Oslo Auroral THEMIS (OATH) data set of all‐sky images, both in terms of runtime and their features' predictive capability. From the features extracted by the best network, we retrain the last neural network layer using the Support Vector Machine (SVM) algorithm to distinguish between the labels “arc,” “diffuse,” “discrete,” “cloud,” “moon” and “clear sky/ no aurora”. This transfer learning approach yields 73% accuracy in the six classes; if we aggregate the 3 auroral and 3 non‐aurora classes, we achieve up to 91% accuracy. We apply our classifier to a new dataset of 550,000 images and evaluate the classifier based on these previously unseen images. To show the potential usefulness of our feature extractor and classifier, we investigate two test cases: First, we compare our predictions for the “cloudy” images to meteorological data and second we train a linear ridge model to predict perturbations in Earth's locally measured magnetic field. We demonstrate that the classifier can be used as a filter to remove cloudy images from datasets and that the extracted features allow to predict magnetometer measurements. All procedures and algorithms used in this study are publicly available, and the code and classifier are provided, which opens possibility for large scale studies of all‐sky images.
