Deep convolutional neural networks (CNNs) achieve state-of-the-art performance in wearable human activity recognition (HAR), which has become a new research trend ubiquitous computing scenario. Increasing network depth or width can further improve accuracy. However, order to obtain the optimal HAR on mobile platform, it consider reasonable tradeoff between accuracy and resource consumption. Improving of CNNs without increasing memory computational burden is more beneficial for HAR. In this...

Efficiently identifying activities of daily living (ADL) provides very important contextual information that is able to improve the effectiveness various sports tracking and healthcare applications. Recently, attention mechanism selectively focuses on time series signals has been widely adopted in sensor based human activity recognition (HAR), which can enhance interesting target ignore irrelevant background activity. Several mechanisms have investigated, achieve remarkable performance HAR...

Recently, convolutional neural networks (CNNs) have set latest state-of-the-art on various human activity recognition (HAR) datasets. However, deep CNNs often require more computing resources, which limits their applications in embedded HAR. Although many successful methods been proposed to reduce memory and FLOPs of CNNs, they involve special network architectures designed for visual tasks, are not suitable HAR tasks with time series sensor signals, due remarkable discrepancy. Therefore, it...

Recently, deep learning has represented an important research trend in human activity recognition (HAR). In particular, convolutional neural networks (CNNs) have achieved state-of-the-art performance on various HAR datasets. For learning, improvements to heavily rely increasing model size or capacity scale larger and datasets, which inevitably leads the increase of operations. A high number operations leaning increases computational cost is not suitable for real-time using mobile wearable...

During recent years, channel attention has attracted great interest in deep learning community. Despite significant success, it been rarely exploited ubiquitous human activity recognition (HAR) scenario. To decrease computational overhead, the often uses global averaging pooling (GAP) to compress each into a simple scalar. It is well known that GAP equal lowest frequency component. obvious lightweight advantage, such compression process inevitably causes severe information loss. In this...

Nowadays, advanced building envelopes not only need to meet traditional design requirements but also address emerging demands, such as achieving low-carbon transition of buildings and mitigating the urban heat island (UHI) effect. Given intricacy indoor conditions complexity variables, approaches can hardly keep pace with evolving demands. Therefore, integrating Artificial Intelligence (AI) into envelope is trending in recent years. This paper provides a holistic review research on machine...

The popularity of convolutional architecture has made sensor-based human activity recognition (HAR) become one primary beneficiary. By simply superimposing multiple convolution layers, the local features can be effectively captured from multi-channel time series sensor data, which could output high-performance prediction results. On other hand, recent years have witnessed great success Transformer model, uses powerful self-attention mechanism to handle long-range sequence modeling tasks,...

Sensor-based human activity recognition (HAR) can provide users with more convenience and security than machine vision-based methods. The ideal HAR application would achieve high-accuracy low-latency performance minimal hardware consumption. However, so far, the speed-accuracy tradeoff in mobile device-based has rarely been systematically analyzed discussed. In contrast to static networks of existing deep models, this article aims design a resource-efficient dynamic network, called RepHAR,...

Graph Neural Networks (GNNs) have significant advantages in handling non-Euclidean data and been widely applied across various areas, thus receiving increasing attention recent years. The framework of GNN models mainly includes the information propagation phase aggregation phase, treating nodes edges as entities channels, respectively. However, most existing face challenge disconnection between node edge feature information, these typically treat learning features independent tasks. To...

Sensor-based human activity recognition (HAR) has become an important task in various application domains. However, existing HAR practices such as convolutional networks and recurrent architectures can result information loss when considering the temporal sensor modality dimensions separately. In this article, we propose a shallow large receptive field (LRF) attention that addresses issue by extracting both temporal-wise modality-wise for sensor-based scenarios. Our proposed LRF architecture...

Recently, human activity recognition (HAR) has become an active research area in wearable computing scenario. On the other hand, residual nets have continued to push state-of-the-art of computer vision and natural language processing. However, rarely been considered HAR field. As grow deeper, memory footprint limit its wide use for a variety tasks. In this paper, we present novel block-wise training that local loss functions applications. Instead global backprop, cross-entropy together with...

Unlike image data, it is often hard to understand intricate sensor data for human activity, which generally contains heterogeneous modalities from different body positions. The importance of every modality might also vary over time. Recent studies have witnessed the success channel attention in boosting model performance. To maintain considerably low computational overhead, utilizes global pooling operation squeeze information, but neglects temporal-aware and modality-aware information that...

As a general model compression paradigm, feature-based knowledge distillation allows the student to learn expressive features from teacher counterpart. In this paper, we mainly focus on designing an effective feature-distillation framework and propose spatial-channel adaptive masked (AMD) network for object detection. More specifically, in order accurately reconstruct important feature regions, first perform attention-guided masking map of network, such that can identify via spatially...

This paper mainly analyzes the performance degradation of turbomachinery in gas turbines, classifies main types degradation: increased tip clearance, corrosion/wear, fouling, and multiple degradation, predicts trend through deep neural networks. The feedforward network is used to build regression model two classification models. uses a back propagation algorithm optimized by Lenvenberg Marquardt convert efficiency flow capacity calculated thermodynamic into values under full load ISO...

Recent mainstream masked distillation methods function by reconstructing selectively areas of a student network from the feature map its teacher counterpart. In these methods, regions need to be properly selected, such that reconstructed features encode sufficient discrimination and representation capability like feature. However, previous only focus on spatial masking, making resulting biased towards importance without encoding informative channel clues. this study, we devise Dual Masked...

<p><em>During training process of LSTM, the prediction accuracy is affected by a variation factors, including selection samples, network structure, optimization algorithm, and stock market status. This paper tries to conduct systematic research on several influencing factors LSTM in context time series prediction. The experiment uses Shanghai Shenzhen 300 constituent stocks from 2006 2017 as samples. study include indicator sampling, sample length, method, data bull bear market,...

This paper introduces a new idea FFCBS, on time series distance measurement, which cares the cost that transforms one to another one. The operation of amplitude scaling, y-axis shifting, are introduced into our give simple definition. Furthermore, c fast warping method is proposed deal with dyerent different lengths, make FFCBS more applicable. With these set-up models. extrapolation solution given, followed by discussion several issues. Experimental result shows approach rather and reasonable.