Title End-to-end eye-movement event detection using deep neural networks /
Authors Zemblys, Raimondas ; Diederick, Niehorster ; Holmqvist, Kenneth
DOI 10.16910/jemr.10.6
Full Text Download
Is Part of Journal of Eye Movement Research: vol. 10, iss. 6: Abstracts of the 19th European Conference on Eye Movements, August 20-24, 2017, Wuppertal, Germany.. Bern : International Group for Eye Movement Research. 2017, vol. 10, iss. 6, p. 151.. ISSN 1995-8692
Keywords [eng] eye-movements ; event detection ; neural networks ; deep learning
Abstract [eng] Existing event detection algorithms for eye-movement data almost exclusively rely on thresholding one or more hand-crafted signal features, each computed from the stream of raw gaze data. Moreover, this thresholding is usually left for the end user. Zemblys et al (2017) present an event detector based on Random Forests, where they show how to train a computationally inexpensive classifier to produce oculomotor events, without the need for a user to set any parameters. This approach outperformed conventional event detection algorithms, approaching the accuracy of expert human coders. However, Random Forests and other traditional machine learning algorithms still need a collection of hand-crafted data descriptors and signal processing features. In this paper, we take one step further and use an end-to-end deep learning approach to classify raw gaze data into fixations, saccades and PSOs. Our method challenges an established tacit assumption that hand-crafted features are necessary in the design of event detection algorithms. Using manually or algorithmically coded examples, we train a LSTM neural network that produces meaningful eye-movement event classification from raw eye-movement data without any need for pre- and post-processing steps. Its accuracy is also at the level of expert human coders.
Published Bern : International Group for Eye Movement Research
Type Conference paper
Language English
Publication date 2017
CC license CC license description