Our experiments show that the model quickly converges to the optimal number of features in a large-scale online setting, and outperforms the (non-incremental) denoising autoencoder, as well as deep belief networks and stacked denoising autoencoders for classification tasks. Specifically, it adds new features to minimize the objective function’s residual and merges similar features to obtain a compact feature representation and prevent over-fitting. This algorithm is composed of two processes: adding features and merging features. In this paper, we propose an incremental feature learning algorithm to determine the optimal model complexity for large-scale, online datasets based on the denoising autoencoder. %X While determining model complexity is an important problem in machine learning, many feature learning algorithms rely on cross-validation to choose an optimal number of features, which is usually infeasible for online learning from a massive stream of data. %C Proceedings of Machine Learning Research %B Proceedings of the Fifteenth International Conference on Artificial Intelligence and Statistics %T Online Incremental Feature Learning with Denoising Autoencoders Further, the algorithm is particularly effective in recognizing new patterns when the data distribution changes over time in the massive online data stream.Ĭite this = Zitt, J., Paitz, P., Walter, F., and Umlauft, J.: Denoising Cryoseismological Distributed Acoustic Sensing Data Using a Deep Neural Network, EGU General Assembly 2023, Vienna, Austria, 24–, EGU23-13269,, 2023.While determining model complexity is an important problem in machine learning, many feature learning algorithms rely on cross-validation to choose an optimal number of features, which is usually infeasible for online learning from a massive stream of data. Finally, suitability, performance as well as advantages and disadvantages of the different types of training data are discussed. To investigate the autoencoder’s general suitability and performance, three different types of training data are tested: purely synthetic data, original data from on-site seismometers, and original data from the DAS recordings themselves. We test this approach on the continuous microseismic Rhonegletscher DAS records. An autoencoder can potentially separate the incoherent noise (such as wind or water flow) from the temporally and spatially coherent signals of interest (e.g., stick-slip event or crevasse formation). Here we propose an autoencoder, which is a deep neural network, as a denoising tool for the analysis of our cryospheric seismic data. Therefore, new techniques are required to denoise the data efficiently and to unmask the signals of interest. Due to the highly active and dynamic cryospheric environment, our collected DAS data are characterized by a low signal to noise ratio compared to classical point sensors. During one month 17 TB of data were acquired. The data collection took place in July 2020 on Rhonegletscher, Switzerland, where a 9 km long fiber-optic cable was installed, covering the entire glacier from its accumulation to its ablation zone. We utilized a DAS unit in a cryospheric environment on a temperate glacier. This technology enables researchers to acquire seismic monitoring data on poorly accessible terrain with great spatial and temporal resolution. Those signals potentially stay unnoticed and thus, might not be analyzed further.ĭistributed acoustic sensing (DAS) is an emerging technology for measuring strain rate data by using common fiber-optic cables in combination with an interrogation unit. One major challenge in Environmental Seismology is that signals of interest are often buried within the high noise level emitted by a multitude of environmental processes.
0 Comments
Leave a Reply. |