Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session B: Wednesday, August 13, 1:00 – 4:00 pm, de Brug & E‑Hall

Toward Generating Plausible Artificial Electroencephalography Data: Evaluating the Effects of Convolution-based Upsampling Methods on Data Quality

Clemens Kaiser¹, Natasha Maurits, Joukje van der Naalt, Marieke van Vugt; ¹University of Groningen

Presenter: Clemens Kaiser

Electroencephalography is key for clinical and cognitive research, yet limited data availability restricts deep learning (DL) applications. This study compares thee upsampling techniques in convolution-based Generative Adversarial Networks {--} transposed convolutions (TC), interpolation with convolutions (IC), and a mixed approach {--} that are used to transform noise vectors into artificial EEG data. We evaluate artificial signal quality with EEG-specific metrics across time, frequency, and spatial domains. Kolmogorov–Smirnov tests indicate that the mixed approach mitigates the high-frequency noise commonly introduced by TC, while better preserving lower-frequency components and inter-channel dependencies than IC. Moreover, the findings underline the importance of EEG-specific evaluation metrics for guiding the development of more explainable and efficacious generative models, advancing DL applications in neuroscience.

Topic Area: Brain Networks & Neural Dynamics

Extended Abstract: Full Text PDF