Analyzing polysomnography (PSG) is an effective method for evaluating sleep health; however, the sleep stage scoring required for PSG analysis is a time-consuming effort for an experienced medical expert. When scoring sleep epochs, experts pay attention to find specific signal characteristics (e.g., K-complexes and spindles), and sometimes need to integrate information from preceding and subsequent epochs in order to make a decision. To imitate this process and to build a more interpretable deep learning model, we propose a neural network based on a convolutional network (CNN) and attention mechanism to perform automatic sleep staging. The CNN learns local signal characteristics, and the attention mechanism excels in learning inter- and intra-epoch features. In experiments on the public sleep-edf and sleep-edfx databases with different training and testing set partitioning methods, our model achieved overall accuracies of 93.7% and 82.8%, and macro-average F1-scores of 84.5 and 77.8, respectively, outperforming recently reported machine learning-based methods.

Primary cilia frequency and length are key metrics in studies of ciliogenesis and ciliopathies. Typically, quantitative cilia analysis is done manually, which is very time-consuming. While some open-source and commercial image analysis software applications can segment input data, they still require the user to optimize many parameters, suffer from user bias, and often lack rigorous performance quality assessment (e.g., false positives and false negatives). Further, optimal parameter combinations vary in detection accuracy depending on cilia reporter, cell type, and imaging modality. A good automated solution would analyze images quickly, robustly, and adaptably—across different experimental data sets—without significantly compromising the accuracy of manual analysis. To solve this problem, we developed a new software for automated cilia detection in cells (ACDC). The software operates through four main steps: image importation, pre-processing, detection auto-optimization, and analysis. From a data set, a representative image with manually selected cilia (i.e., Ground Truth) is used for detection auto-optimization based on four parameters: signal-to-noise ratio, length, directional score, and intensity standard deviation. Millions of parameter combinations are automatically evaluated and optimized according to an accuracy ‘F1’ score, based on the amount of false positives and false negatives. Afterwards, the optimized parameter combination is used for automated detection and analysis of the entire data set. The ACDC software accurately and adaptably detected nuclei and primary cilia across different cell types (NIH3T3, RPE1), cilia reporters (AcTub, Smo-GFP, Arl13b), and image magnifications (60×, 40×). We found that false-positive and false-negative rates for Arl13b-stained cilia were 1–6%, yielding high F1 scores of 0.96–0.97 (max. = 1.00). The software detected significant differences in mean cilia length between control and cytochalasin D-treated cell populations and could monitor dynamic changes in cilia length from movie recordings. Automated analysis offered up to a 96-fold speed enhancement compared to manual analysis, requiring around 5 s/image, or nearly 18,000 cilia analyzed/hour. The ACDC software is a solution for robust automated analysis of microscopic images of ciliated cells. The software is extremely adaptable, accurate, and offers immense time-savings compared to traditional manual analysis.

Click-Through Rate (CTR) prediction is a crucial task for various online applications, such as recommendation and online advertising. The task of CTR prediction is to predict the probability of users' clicking behaviors, with high-dimensional input features. To avoid heavy handcrafted feature engineering, the core topic of CTR prediction is the automatic interactions of the input features. Factorization Machine (FM) is an effective approach for modeling second-order feature interactions. Recently, FM has been extended for modeling higher-order feature interactions, such as xDeepFM and Higher-Order Factorization Machine (HOFM). However, these approaches are with either high complexity or iterative computation consuming much time and space. To overcome above problems, we express arbitrary-order FM in the form of power sums according to Newton's identities. Accordingly, we propose a novel Interaction Machine (IM) model. IM is an efficient and exact implementation of high-order FM, whose time complexity linearly grows with the order of interactions and the number of feature fields. Via IM, we can conduct arbitrary-order feature interactions in a very simple way. Moreover, we perform IM together with deep neural networks, and the resulted DeepIM model is more efficient than xDeepFM with comparable or even better performance. We conduct experiments on two real-world datasets, in which effectiveness and efficiency of both IM and DeepIM are strongly verified.

We present a novel pipelined fast Fourier transform (FFT) architecture which is capable of producing the output sequence in normal order. A single-path delay commutator processing element (SDC PE) has been proposed for the first time. It saves a complex adder compared with the typical radix-2 butterfly unit. The new pipelined architecture can be built using the proposed processing element. The proposed architecture can lead to 100% hardware utilization and 50% reduction in the overall number of adders required in the conventional pipelined FFT designs. In order to produce the output sequence in normal order, we also present a bit reverser, which can achieve a 50% reduction in memory usage.

Electroencephalography (EEG) signals are prone to contamination by noise, such as ocular and muscle artifacts. Minimizing these artifacts is crucial for EEG-based downstream applications like disease diagnosis and brain-computer interface (BCI). This paper presents a new EEG denoising model, DTP-Net. It is a fully convolutional neural network comprising Densely-connected Temporal Pyramids (DTPs) placed between two learnable time-frequency transformations. In the time-frequency domain, DTPs facilitate efficient propagation of multi-scale features extracted from EEG signals of any length, leading to effective noise reduction. Comprehensive experiments on two public semi-simulated datasets demonstrate that the proposed DTP-Net consistently outperforms existing state-of-the-art methods on metrics including relative root mean square error (RRMSE) and signal-to-noise ratio improvement ( $\Delta$ SNR). Moreover, the proposed DTP-Net is applied to a BCI classification task, yielding an improvement of up to 5.55% in accuracy. This confirms the potential of DTP-Net for applications in the fields of EEG-based neuroscience and neuro-engineering. An in-depth analysis further illustrates the representation learning behavior of each module in DTP-Net, demonstrating its robustness and reliability.

We present a hybrid pipelined sorting architecture capable of finding and producing as its output the K largest elements from an input sequence. The architecture consists of a bitonic sorter and L cascaded sorting units. The sorting unit is designed to output P elements during every cycle with the aim of increasing the throughput and lowering the latency. The function of the bitonic sorter is to generate a segmented ordered sequence. The sorting unit processes this sequence to identify and output the P largest elements. Hence, the K=PL largest elements are obtained after the segmented ordered sequence proceeds through L cascaded sorting units. Variable-length and continuous sequences are supported by the proposed sorting architecture. The results of the implementation show that the sorting architecture can achieve a throughput of 22.88 GB/s with P=16 on a state-of-the-art Field Programmable Gate Array (FPGA).