EEG-EyeTrack: A Benchmark for Time Series and Functional Data Analysis with Open Challenges and Baselines

TL;DR

A new benchmark dataset for functional data analysis (FDA) is presented, focusing on the reconstruction of eye movements from EEG data, and functional neural networks are used to establish baseline results for the primary regression task of reconstructing eye movements from EEG signals.

Abstract

A new benchmark dataset for functional data analysis (FDA) is presented, focusing on the reconstruction of eye movements from EEG data. The contribution is twofold: first, open challenges and evaluation metrics tailored to FDA applications are proposed. Second, functional neural networks are used to establish baseline results for the primary regression task of reconstructing eye movements from EEG signals. Baseline results are reported for the new dataset, based on consumer-grade hardware, and the EEGEyeNet dataset, based on research-grade hardware.

Figures (2)

• Figure 1: Comparison of various neural networks on the level-1 saccades task. Every prediction made on the test set is shown as a dot. The dots are colored based on the true target position. The four target positions are indicated by black crosses.
• Figure 2: The architecture of the SpatialFilterCNN model, which on a high level consists of the input, a spatial filtering layer, two residual blocks and a fully connected neural network at the output. Conv1D($n$, $m$) denotes a one-dimensional convolutional layer with $n$ filters and a kernel size of $m$, AvgPooling($n$, $m$) average pooling with pool size $n$ and stride $m$ and Linear($n$) a layer of $n$ fully connected neurons.