Surrogate Data for Deep Learning Architectures in Rehabilitative Edge Systems

Lee, T. K. M.; Chan, H. W.; Leo, K. H.; Chew, Effie; Zhao, Ling; Sanei, Saeid

doi:10.23919/spa50552.2020.9241275

Cited by 2 publications

(2 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Schwabedal et al [40] adopted a similar approach, but applied surrogates only to the training data, while performing cross-validation. On the other hand, in a later study, Lee et al [41] proposed using surrogates as the exclusive basis for training and validation, reserving the entirety of the experimental recordings for testing. In this paper, we consider three possibilities more systematically: training only on the original data, training only on the surrogates, and training on an evenly mixed pool.…”

Section: Related Workmentioning

confidence: 99%

“…Importantly, this also counters the risk of overfitting. Another innovation is that, unlike the existing studies such as the one by Lee et al [41], we explicitly consider the multivariate nature of the triaxial vector accelerometer data and consequently employ an iterative method that preserves not only the autocorrelations but also the crosscorrelations, more appropriately representing the real-world kinematics. Finally, the authors are unaware of any studies that explicitly attempt to demonstrate why surrogate data help for data augmentation.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Integrated Data Augmentation for Accelerometer Time Series in Behavior Recognition: Roles of Sampling, Balancing, and Fourier Surrogates

Minati

Tokgoz

et al. 2022

IEEE Sensors J.

View full text Add to dashboard Cite

The behavioral monitoring of farmed animals such as cattle is a fundamental element of precision farming in that it enables unobtrusive ongoing health monitoring. This application presents two ubiquitous challenges typical of sensing applications of internet-of-things: limited dataset size and dataset imbalance. Recently, data augmentation has emerged as a way of addressing their negative influences on the training process without overburdening the data acquisition phase. However, there remains no consensus regarding which methods should be applied to time series and in what combination. Here, we present the first comprehensive analysis that synergistically combines multiple approaches. These approaches are benchmarked on a dataset of triaxial accelerometer time series, which were acquired from six freely roaming cows through a collar-mounted sensor, and labeled by experienced human observers according to five behaviors. Our results indicate that integrating data augmentation with the training process can substantially improve time series classification performance while retaining a fixed convolutional neural network architecture. The improvement is maximized when the dataset is balanced by applying a suitable sampling scheme and the negative influence of data duplication is reduced via generating synthetic time series with Fourier surrogates. With the proposed approach, the overall accuracy is improved from 90% to 96%, and the classification accuracy of an under-represented behavior, namely grazing, is elevated from 45% to 91%. This work provides a direction toward a general methodology, motivating research on other datasets and applications.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%