A Spectral Enabled GAN for Time Series Data Generation

Smith, Kaleb E.; Smith, Anthony O.

doi:10.48550/arxiv.2103.01904

Cited by 5 publications

(7 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We use 2D Conv for the spectral WGAN and 1D Conv for the temporal WGAN. The loss function and optimizers follow as described in the original work (Smith and Smith, 2020 , 2021 ).…”

Section: Resultsmentioning

confidence: 99%

E2SGAN: EEG-to-SEEG translation with generative adversarial networks

Chen

Jiang

et al. 2022

Front. Neurosci.

View full text Add to dashboard Cite

High-quality brain signal data recorded by Stereoelectroencephalography (SEEG) electrodes provide clinicians with clear guidance for presurgical assessments for epilepsy surgeries. SEEG, however, is limited to selected patients with epilepsy due to its invasive procedure. In this work, a brain signal synthesis framework is presented to synthesize SEEG signals from non-invasive EEG signals. First, a strategy to determine the matching relation between EEG and SEEG channels is presented by considering both signal correlation and spatial distance. Second, the EEG-to-SEEG generative adversarial network (E2SGAN) is proposed to precisely synthesize SEEG data from the simultaneous EEG data. Although the widely adopted magnitude spectra has proved to be informative in EEG tasks, it leaves much to be desired in the setting of signal synthesis. To this end, instantaneous frequency spectra is introduced to further represent the alignment of the signal. Correlative spectral attention (CSA) is proposed to enhance the discriminator of E2SGAN by capturing the correlation between each pair of EEG and SEEG frequencies. The weighted patch prediction (WPP) technique is devised to ensure robust temporal results. Comparison experiments on real-patient data demonstrate that E2SGAN outperforms baseline methods in both temporal and frequency domains. The perturbation experiment reveals that the synthesized results have the potential to capture abnormal discharges in epileptic patients before seizures.

show abstract

“…We use 2D Conv for the spectral WGAN and 1D Conv for the temporal WGAN. The loss function and optimizers follow as described in the original work (Smith and Smith, 2020 , 2021 ).…”

Section: Resultsmentioning

confidence: 99%

E2SGAN: EEG-to-SEEG translation with generative adversarial networks

Chen

Jiang

et al. 2022

Front. Neurosci.

View full text Add to dashboard Cite

show abstract

“…Distributing TimeGAN computations is complicated, which leads to long training procedures. Other approaches use convolutional neural networks to simulate image representations of time series (Donahue et al, 2019) (Smith and Smith, 2021) and are currently dominated by the use of spectrograms and waveforms. Spectrograms process audio signals and visualize the spectrum of frequencies of a signal as it varies with time (Wyse, 2017) while each plot is characteristic for a certain sound or a spoken word.…”

Section: Related Workmentioning

confidence: 99%

“…Spectrograms process audio signals and visualize the spectrum of frequencies of a signal as it varies with time (Wyse, 2017) while each plot is characteristic for a certain sound or a spoken word. In a recent paper, Smith and Smith (2021) propose a GAN model with a simplified, image-based approach using two separate Wasserstein GANs. The first network takes the spectral image of the signal and generates new spectrograms.…”

Section: Related Workmentioning

confidence: 99%

“…Contrary to the CNN-based models used for image data generation, RNN-based time series GAN models are often complex and consist of numerous networks, encoding/decoding steps, and loss functions. Inspired by recent successes in computer vision, multiple attempts have been made to simulate time series using convolutional neural networks (Brophy et al, 2019;Smith and Smith, 2021). However, these approaches lack a proper image representation of the time series.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Leveraging Image-based Generative Adversarial Networks for Time Series Generation

Hellermann¹,

Lessmann²

2021

Preprint

View full text Add to dashboard Cite

Generative models synthesize image data with great success regarding sampling quality, diversity and feature disentanglement. Generative models for time series lack these benefits due to a missing representation, which captures temporal dynamics and allows inversion for sampling. The paper proposes the intertemporal return plot (IRP) representation to facilitate the use of image-based generative adversarial networks for time series generation. The representation proves effective in capturing time series characteristics and, compared to alternative representations, benefits from invertibility and scale-invariance. Empirical benchmarks confirm these features and demonstrate that the IRP enables an off-the-shelf Wasserstein GAN with gradient penalty to sample realistic time series, which outperform a specialized RNN-based GAN, while simultaneously reducing model complexity.

show abstract

“…This image processing has been the starting point when using GAN networks in other types of applications, such as 1D signal processing. The change from 1D (time signals) to 2D (images) is achieved by generating spectrograms or scalograms with the input signal applying a Short-Time Fourier Transform (STFT) [7]. This change of dimensions is not specific to GANs, it is also performed on other topologies such as U-Net (e.g.…”

Section: Introductionmentioning

confidence: 99%