Low-dimensional Convolutional Neural Network for Solar Flares GOES Time-series Classification

Space weather phenomena such as solar flares have a massive destructive power when they reach a certain magnitude. Here, we explore the deep-learning approach in order to build a solar flare-forecasting model, while examining its limitations and feature-extraction ability based on the available Geostationary Operational Environmental Satellite (GOES) X-ray time-series data. We present a multilayer 1D convolutional neural network to forecast the solar flare event probability occurrence of M- and X-class flares at 1, 3, 6, 12, 24, 48, 72, and 96 hr time frames. The forecasting models were trained and evaluated in two different scenarios: (1) random selection and (2) chronological selection, which were compared afterward in terms of common score metrics. Additionally, we also compared our results to state-of-the-art flare-forecasting models. The results indicates that (1) when X-ray time-series data are used alone, the suggested model achieves higher score results for X-class flares and similar scores for M-class as in previous studies. (2) The two different scenarios obtain opposite results for the X- and M-class flares. (3) The suggested model combined with solely X-ray time-series fails to distinguish between M- and X-class magnitude solar flare events. Furthermore, based on the suggested method, the achieved scores, obtained solely from X-ray time-series measurements, indicate that substantial information regarding the solar activity and physical processes are encapsulated in the data, and augmenting additional data sets, both spatial and temporal, may lead to better predictions, while gaining a comprehensive physical interpretation regarding solar activity. All source codes are available at https://github.com/vladlanda.

A Framework for Ultra Low-Power Hardware Accelerators Using NNs for Embedded Time Series Classification

In embedded applications that use neural networks (NNs) for classification tasks, it is important to not only minimize the power consumption of the NN calculation, but of the whole system. Optimization approaches for individual parts exist, such as quantization of the NN or analog calculation of arithmetic operations. However, there is no holistic approach for a complete embedded system design that is generic enough in the design process to be used for different applications, but specific in the hardware implementation to waste no energy for a given application. Therefore, we present a novel framework that allows an end-to-end ASIC implementation of a low-power hardware for time series classification using NNs. This includes a neural architecture search (NAS), which optimizes the NN configuration for accuracy and energy efficiency at the same time. This optimization targets a custom designed hardware architecture that is derived from the key properties of time series classification tasks. Additionally, a hardware generation tool is used that creates a complete system from the definition of the NN. This system uses local multi-level RRAM memory as weight and bias storage to avoid external memory access. Exploiting the non-volatility of these devices, such a system can use a power-down mode to save significant energy during the data acquisition process. Detection of atrial fibrillation (AFib) in electrocardiogram (ECG) data is used as an example for evaluation of the framework. It is shown that a reduction of more than 95% of the energy consumption compared to state-of-the-art solutions is achieved.

Time Series Classification with InceptionFCN

Deep neural networks (DNN) have proven to be efficient in computer vision and data classification with an increasing number of successful applications. Time series classification (TSC) has been one of the challenging problems in data mining in the last decade, and significant research has been proposed with various solutions, including algorithm-based approaches as well as machine and deep learning approaches. This paper focuses on combining the two well-known deep learning techniques, namely the Inception module and the Fully Convolutional Network. The proposed method proved to be more efficient than the previous state-of-the-art InceptionTime method. We tested our model on the univariate TSC benchmark (the UCR/UEA archive), which includes 85 time-series datasets, and proved that our network outperforms the InceptionTime in terms of the training time and overall accuracy on the UCR archive.

Comparing End-to-End Machine Learning Methods for Spectra Classification

In scientific research, spectroscopy and diffraction experimental techniques are widely used and produce huge amounts of spectral data. Learning patterns from spectra is critical during these experiments. This provides immediate feedback on the actual status of the experiment (e.g., time-resolved status of the sample), which helps guide the experiment. The two major spectral changes what we aim to capture are either the change in intensity distribution (e.g., drop or appearance) of peaks at certain locations, or the shift of those on the spectrum. This study aims to develop deep learning (DL) classification frameworks for one-dimensional (1D) spectral time series. In this work, we deal with the spectra classification problem from two different perspectives, one is a general two-dimensional (2D) space segmentation problem, and the other is a common 1D time series classification problem. We focused on the two proposed classification models under these two settings, the namely the end-to-end binned Fully Connected Neural Network (FCNN) with the automatically capturing weighting factors model and the convolutional SCT attention model. Under the setting of 1D time series classification, several other end-to-end structures based on FCNN, Convolutional Neural Network (CNN), ResNets, Long Short-Term Memory (LSTM), and Transformer were explored. Finally, we evaluated and compared the performance of these classification models based on the High Energy Density (HED) spectra dataset from multiple perspectives, and further performed the feature importance analysis to explore their interpretability. The results show that all the applied models can achieve 100% classification confidence, but the models applied under the 1D time series classification setting are superior. Among them, Transformer-based methods consume the least training time (0.449 s). Our proposed convolutional Spatial-Channel-Temporal (SCT) attention model uses 1.269 s, but its self-attention mechanism performed across spatial, channel, and temporal dimensions can suppress indistinguishable features better than others, and selectively focus on obvious features with high separability.